diff --git a/.gitignore b/.gitignore
index 3918d97..0c1de92 100644
--- a/.gitignore
+++ b/.gitignore
@@ -43,4 +43,5 @@ next-env.d.ts
 package-lock.json
 .next/
 
-.venv/
\ No newline at end of file
+.venv/
+roadcast/data.csv
diff --git a/roadcast/README.md b/roadcast/README.md
new file mode 100644
index 0000000..66195d6
--- /dev/null
+++ b/roadcast/README.md
@@ -0,0 +1,74 @@
+# RoAdCast - Flask + PyTorch CNN starter
+
+This project contains a minimal Flask app and a small PyTorch CNN scaffold so you can train and run a model directly from VS Code.
+
+Quick setup (Windows PowerShell):
+
+1. Create and activate a virtual environment
+
+```powershell
+python -m venv .venv
+.\.venv\Scripts\Activate.ps1
+```
+
+2. Install dependencies
+
+```powershell
+pip install -r requirements.txt
+```
+
+
+3. Dataset layout
+
+Image dataset (folder-per-class):
+
+data/
+  class1/
+    img1.jpg
+  class2/
+    img2.jpg
+
+CSV dataset (single file):
+
+data.csv  (expects a `label` column and numeric feature columns)
+
+Train commands:
+
+Image training (default cnn):
+
+```powershell
+python train.py data --epochs 5 --batch-size 16
+```
+
+CSV/tabular training (MLP):
+
+```powershell
+python train.py data.csv --model-type mlp --epochs 20 --batch-size 64
+```
+
+The model will be saved as `model.pth` in the repo root (best validation checkpoint).
+
+Run the Flask app (for local testing):
+
+```powershell
+python app.py
+```
+
+Predict using curl (or Postman). Example with curl in PowerShell:
+
+```powershell
+curl -X POST -F "image=@path\to\image.jpg" http://127.0.0.1:5000/predict
+```
+
+VS Code tips
+
+- Open this folder in VS Code.
+- Use the Python extension and select the `.venv` interpreter.
+- Use the Run panel to add a launch configuration that runs `app.py` or `train.py`.
+- For long training runs, run training in the terminal (not the debugger) and monitor logs.
+
+Notes & next steps
+
+- The SimpleCNN uses 224x224 input and expects at least two maxpool steps; adjust `models.py` if you want smaller inputs.
+- Add better transforms and augmentation in `data.py` for better performance.
+- If GPU is available, PyTorch will use it automatically if installed with CUDA.
diff --git a/roadcast/__pycache__/app.cpython-312.pyc b/roadcast/__pycache__/app.cpython-312.pyc
new file mode 100644
index 0000000..2fae411
Binary files /dev/null and b/roadcast/__pycache__/app.cpython-312.pyc differ
diff --git a/roadcast/__pycache__/data.cpython-312.pyc b/roadcast/__pycache__/data.cpython-312.pyc
new file mode 100644
index 0000000..82122df
Binary files /dev/null and b/roadcast/__pycache__/data.cpython-312.pyc differ
diff --git a/roadcast/__pycache__/inference.cpython-312.pyc b/roadcast/__pycache__/inference.cpython-312.pyc
new file mode 100644
index 0000000..1d75c20
Binary files /dev/null and b/roadcast/__pycache__/inference.cpython-312.pyc differ
diff --git a/roadcast/__pycache__/models.cpython-312.pyc b/roadcast/__pycache__/models.cpython-312.pyc
new file mode 100644
index 0000000..25ec296
Binary files /dev/null and b/roadcast/__pycache__/models.cpython-312.pyc differ
diff --git a/roadcast/__pycache__/openweather_inference.cpython-312.pyc b/roadcast/__pycache__/openweather_inference.cpython-312.pyc
new file mode 100644
index 0000000..ac86aa1
Binary files /dev/null and b/roadcast/__pycache__/openweather_inference.cpython-312.pyc differ
diff --git a/roadcast/app.py b/roadcast/app.py
new file mode 100644
index 0000000..38d3889
--- /dev/null
+++ b/roadcast/app.py
@@ -0,0 +1,123 @@
+from flask import Flask, request, jsonify
+
+app = Flask(__name__)
+import os
+import threading
+import json
+
+# ML imports are lazy to avoid heavy imports on simple runs
+
+
+@app.route('/get-data', methods=['GET'])
+def get_data():
+    # Example GET request handler
+    data = {"message": "Hello from Flask!"}
+    return jsonify(data)
+
+@app.route('/post-data', methods=['POST'])
+def post_data():
+    # Example POST request handler
+    content = request.json
+    # Process content or call AI model here
+    response = {"you_sent": content}
+    return jsonify(response)
+
+
+@app.route('/train', methods=['POST'])
+def train_endpoint():
+    """Trigger training. Expects JSON: {"data_root": "path/to/data", "epochs": 3}
+    Training runs in a background thread and saves model to model.pth in repo root.
+    """
+    payload = request.json or {}
+    data_root = payload.get('data_root')
+    epochs = int(payload.get('epochs', 3))
+    if not data_root or not os.path.isdir(data_root):
+        return jsonify({"error": "data_root must be a valid directory path"}), 400
+
+    def _run_training():
+        from train import train
+        train(data_root, epochs=epochs)
+
+    t = threading.Thread(target=_run_training, daemon=True)
+    t.start()
+    return jsonify({"status": "training_started"})
+
+
+@app.route('/predict', methods=['POST'])
+def predict_endpoint():
+    """Predict single uploaded image. Expects form-data with file field named 'image'."""
+    if 'image' not in request.files:
+        return jsonify({"error": "no image uploaded (field 'image')"}), 400
+    img = request.files['image']
+    tmp_path = os.path.join(os.getcwd(), 'tmp_upload.jpg')
+    img.save(tmp_path)
+    try:
+        from inference import load_model, predict_image
+        model_path = os.path.join(os.getcwd(), 'model.pth')
+        if not os.path.exists(model_path):
+            return jsonify({"error": "no trained model found (run /train first)"}), 400
+        model, idx_to_class = load_model(model_path)
+        idx, conf = predict_image(model, tmp_path)
+        label = idx_to_class.get(idx) if idx_to_class else str(idx)
+        return jsonify({"label": label, "confidence": conf})
+    finally:
+        try:
+            os.remove(tmp_path)
+        except Exception:
+            pass
+
+
+@app.route('/predict-roadrisk', methods=['POST'])
+def predict_roadrisk():
+    """Proxy endpoint to predict a roadrisk cluster from lat/lon/datetime.
+
+    Expects JSON body with: {"lat": 38.9, "lon": -77.0, "datetime": "2025-09-27T12:00:00", "roadrisk_url": "https://..."}
+    If roadrisk_url is not provided the endpoint will call OpenWeather OneCall (requires API key via OPENWEATHER_KEY env var).
+    """
+    payload = request.json or {}
+    lat = payload.get('lat')
+    lon = payload.get('lon')
+    dt = payload.get('datetime')
+    street = payload.get('street', '')
+    roadrisk_url = payload.get('roadrisk_url')
+    api_key = payload.get('api_key') or os.environ.get('OPENWEATHER_KEY')
+
+    if lat is None or lon is None:
+        return jsonify({"error": "lat and lon are required fields"}), 400
+
+    try:
+        from openweather_inference import predict_from_openweather
+        res = predict_from_openweather(lat, lon, dt_iso=dt, street=street, api_key=api_key, train_csv=os.path.join(os.getcwd(), 'data.csv'), preprocess_meta=None, model_path=os.path.join(os.getcwd(), 'model.pth'), centers_path=os.path.join(os.getcwd(), 'kmeans_centers_all.npz'), roadrisk_url=roadrisk_url)
+        return jsonify(res)
+    except Exception as e:
+        return jsonify({"error": str(e)}), 500
+
+
+@app.route('/health', methods=['GET'])
+def health():
+    """Return status of loaded ML artifacts (model, centers, preprocess_meta)."""
+    try:
+        from openweather_inference import init_inference
+        status = init_inference()
+        return jsonify({'ok': True, 'artifacts': status})
+    except Exception as e:
+        return jsonify({'ok': False, 'error': str(e)}), 500
+
+if __name__ == '__main__':
+    # eager load model/artifacts at startup (best-effort)
+    try:
+        from openweather_inference import init_inference
+        init_inference()
+    except Exception:
+        pass
+    app.run(debug=True)
+
+# @app.route('/post-data', methods=['POST'])
+# def post_data():
+#     content = request.json
+#     user_input = content.get('input')
+
+#     # Example: Simple echo AI (replace with real AI model code)
+#     ai_response = f"AI received: {user_input}"
+
+#     return jsonify({"response": ai_response})
diff --git a/roadcast/check_env.py b/roadcast/check_env.py
new file mode 100644
index 0000000..5fd7b84
--- /dev/null
+++ b/roadcast/check_env.py
@@ -0,0 +1,32 @@
+import sys
+import importlib
+
+def safe_import(name):
+    try:
+        return importlib.import_module(name)
+    except Exception as e:
+        return e
+
+print('Python:', sys.version.replace('\n',' '))
+
+torch = safe_import('torch')
+if isinstance(torch, Exception):
+    print('torch import error:', torch)
+else:
+    print('torch:', torch.__version__)
+    print('CUDA available:', torch.cuda.is_available())
+    if torch.cuda.is_available():
+        print('CUDA device count:', torch.cuda.device_count())
+        print('Current device name:', torch.cuda.get_device_name(0))
+
+pandas = safe_import('pandas')
+if isinstance(pandas, Exception):
+    print('pandas import error:', pandas)
+else:
+    print('pandas:', pandas.__version__)
+
+try:
+    import sklearn
+    print('sklearn:', sklearn.__version__)
+except Exception:
+    pass
diff --git a/roadcast/data.py b/roadcast/data.py
new file mode 100644
index 0000000..febbf22
--- /dev/null
+++ b/roadcast/data.py
@@ -0,0 +1,413 @@
+import os
+import hashlib
+from datetime import datetime
+import pandas as pd
+import numpy as np
+import torch
+from torch.utils.data import Dataset
+from PIL import Image
+from torchvision import transforms
+
+
+class ImageFolderDataset(Dataset):
+    """A minimal image folder dataset expecting a structure: root/class_name/*.jpg"""
+    def __init__(self, root, transform=None):
+        self.root = root
+        self.samples = []  # list of (path, label)
+        classes = sorted([d for d in os.listdir(root) if os.path.isdir(os.path.join(root, d))])
+        self.class_to_idx = {c: i for i, c in enumerate(classes)}
+        for c in classes:
+            d = os.path.join(root, c)
+            for fname in os.listdir(d):
+                if fname.lower().endswith(('.png', '.jpg', '.jpeg')):
+                    self.samples.append((os.path.join(d, fname), self.class_to_idx[c]))
+
+        self.transform = transform or transforms.Compose([
+            transforms.Resize((224, 224)),
+            transforms.ToTensor(),
+        ])
+
+    def __len__(self):
+        return len(self.samples)
+
+    def __getitem__(self, idx):
+        path, label = self.samples[idx]
+        img = Image.open(path).convert('RGB')
+        img = self.transform(img)
+        return img, label
+
+
+class CSVDataset(Dataset):
+    """Load classification tabular data from a single CSV file.
+
+    Expects a `label` column and numeric feature columns. Non-numeric columns are dropped.
+    """
+    def __init__(self, csv_path, feature_columns=None, label_column='label', transform=None, generate_labels=False, n_buckets=100, label_method='md5', label_store=None, feature_engineer=False, lat_lon_bins=20, nrows=None):
+        # read CSV with low_memory=False to avoid mixed-type warnings
+        if nrows is None:
+            self.df = pd.read_csv(csv_path, low_memory=False)
+        else:
+            self.df = pd.read_csv(csv_path, nrows=nrows, low_memory=False)
+        self.label_column = label_column
+
+        if generate_labels:
+            # generate deterministic labels based on selected columns
+            self.df[self.label_column] = generate_labels_for_df(self.df, n_buckets=n_buckets, method=label_method, label_store=label_store)
+
+        # optional simple feature engineering: extract date parts and lat/lon bins
+        if feature_engineer:
+            try:
+                _add_date_features(self.df)
+            except Exception:
+                pass
+            try:
+                _add_latlon_bins(self.df, bins=lat_lon_bins)
+            except Exception:
+                pass
+
+        if label_column not in self.df.columns:
+            raise ValueError(f"label column '{label_column}' not found in CSV; set generate_labels=True to create labels")
+
+        # determine feature columns if not provided (numeric columns except label)
+        if feature_columns is None:
+            feature_columns = [c for c in self.df.columns if c != label_column and pd.api.types.is_numeric_dtype(self.df[c])]
+        self.feature_columns = feature_columns
+        # coerce feature columns to numeric, fill NaNs with column mean (or 0), then standardize
+        features_df = self.df[self.feature_columns].apply(lambda c: pd.to_numeric(c, errors='coerce'))
+        # fill NaNs with column mean where possible, otherwise 0
+        initial_means = features_df.mean()
+        features_df = features_df.fillna(initial_means).fillna(0.0)
+
+        # drop columns that remain all-NaN after coercion/fill (unlikely after fillna(0.0)), to avoid NaNs
+        all_nan_cols = features_df.columns[features_df.isna().all()].tolist()
+        if len(all_nan_cols) > 0:
+            # remove from feature list so indices stay consistent
+            features_df = features_df.drop(columns=all_nan_cols)
+            self.feature_columns = [c for c in self.feature_columns if c not in all_nan_cols]
+
+        # recompute means/stds from the filled data so subtraction/division won't produce NaNs
+        col_means = features_df.mean()
+        col_stds = features_df.std().replace(0, 1.0).fillna(1.0)
+
+        # standardize using the recomputed stats
+        features_df = (features_df - col_means) / (col_stds + 1e-6)
+
+        self.feature_means = col_means.to_numpy(dtype=float)
+        self.feature_stds = col_stds.to_numpy(dtype=float)
+
+        self.features = torch.tensor(features_df.values, dtype=torch.float32)
+        self.labels = torch.tensor(pd.to_numeric(self.df[self.label_column], errors='coerce').fillna(0).astype(int).values, dtype=torch.long)
+
+    def __len__(self):
+        return len(self.df)
+
+    def __getitem__(self, idx):
+        return self.features[idx], int(self.labels[idx])
+
+
+def _normalize_str(x):
+    if pd.isna(x):
+        return ''
+    return str(x).strip().lower()
+
+
+def _normalize_date(x):
+    try:
+        # try parse common formats
+        dt = pd.to_datetime(x)
+        return dt.strftime('%Y-%m-%d')
+    except Exception:
+        return ''
+
+
+def generate_kmeans_labels(df, n_buckets=100, random_state=42, label_store=None):
+    """Generate labels by running k-means over numeric features (deterministic with seed).
+
+    This produces clusters that are predictable from numeric inputs and are therefore
+    better suited for training a numeric-feature MLP than arbitrary hash buckets.
+    """
+    # small pure-numpy k-means to avoid external dependency
+    import numpy as np
+
+    # select numeric columns only
+    num_df = df.select_dtypes(include=['number']).fillna(0.0)
+    if num_df.shape[0] == 0 or num_df.shape[1] == 0:
+        # fallback to hashing if no numeric columns
+        return generate_labels_for_df(df, n_buckets=n_buckets)
+
+    data = num_df.values.astype(float)
+    n_samples = data.shape[0]
+    rng = np.random.default_rng(random_state)
+
+    # If a label_store exists and contains centers, load and use them
+    import os
+    if label_store and os.path.exists(label_store):
+        try:
+            npz = np.load(label_store)
+            centers = npz['centers']
+            all_dists = np.linalg.norm(data[:, None, :] - centers[None, :, :], axis=2)
+            all_labels = np.argmin(all_dists, axis=1)
+            return pd.Series(all_labels, index=df.index)
+        except Exception:
+            # fall through to fitting
+            pass
+
+    # sample points to fit centers if dataset is large
+    sample_size = min(20000, n_samples)
+    if sample_size < n_samples:
+        idx = rng.choice(n_samples, size=sample_size, replace=False)
+        sample_data = data[idx]
+    else:
+        sample_data = data
+
+    # initialize centers by random sampling from sample_data
+    centers_idx = rng.choice(sample_data.shape[0], size=min(n_buckets, sample_data.shape[0]), replace=False)
+    centers = sample_data[centers_idx].astype(float)
+
+    # run a small number of iterations
+    max_iters = 10
+    for _ in range(max_iters):
+        # assign
+        dists = np.linalg.norm(sample_data[:, None, :] - centers[None, :, :], axis=2)
+        labels = np.argmin(dists, axis=1)
+        # recompute centers
+        new_centers = np.zeros_like(centers)
+        counts = np.zeros((centers.shape[0],), dtype=int)
+        for i, lab in enumerate(labels):
+            new_centers[lab] += sample_data[i]
+            counts[lab] += 1
+        for k in range(centers.shape[0]):
+            if counts[k] > 0:
+                new_centers[k] = new_centers[k] / counts[k]
+            else:
+                # reinitialize empty cluster
+                new_centers[k] = sample_data[rng.integers(0, sample_data.shape[0])]
+        # check convergence (centers change small)
+        shift = np.linalg.norm(new_centers - centers, axis=1).max()
+        centers = new_centers
+        if shift < 1e-4:
+            break
+
+    # assign labels for all data
+    all_dists = np.linalg.norm(data[:, None, :] - centers[None, :, :], axis=2)
+    all_labels = np.argmin(all_dists, axis=1)
+    # persist centers if requested
+    if label_store:
+        try:
+            np.savez_compressed(label_store, centers=centers)
+        except Exception:
+            pass
+    return pd.Series(all_labels, index=df.index)
+
+
+def generate_labels_for_df(df, n_buckets=100, method='md5', label_store=None):
+    """Generate deterministic bucket labels 1..n_buckets from rows using selected columns.
+
+    Uses: report_dat, latitude, longitude, street1, street2, ward, injuries, fatalities.
+    Produces reproducible labels via md5 hashing of a normalized feature string.
+    """
+    if method == 'kmeans':
+        return generate_kmeans_labels(df, n_buckets=n_buckets, label_store=label_store)
+
+    # Be flexible about column names (case variations and alternate names).
+    colmap = {c.lower(): c for c in df.columns}
+
+    def get_col(*candidates):
+        for cand in candidates:
+            key = cand.lower()
+            if key in colmap:
+                return colmap[key]
+        return None
+
+    report_col = get_col('report_dat', 'reportdate', 'fromdate', 'lastupdatedate')
+    lat_col = get_col('latitude', 'mpdlatitude', 'lat')
+    lon_col = get_col('longitude', 'mpdlongitude', 'lon')
+    street1_col = get_col('street1', 'address', 'mar_address', 'nearestintstreetname')
+    street2_col = get_col('street2', 'nearestintstreetname')
+    ward_col = get_col('ward')
+
+    inj_cols = [c for c in df.columns if 'INJUR' in c.upper()]
+    fat_cols = [c for c in df.columns if 'FATAL' in c.upper()]
+
+    uid = get_col('crimeid', 'eventid', 'objectid', 'ccn')
+
+    def row_to_bucket(row):
+        parts = []
+        # date
+        parts.append(_normalize_date(row.get(report_col, '') if report_col else ''))
+        # lat/lon rounded
+        lat = row.get(lat_col, '') if lat_col else ''
+        lon = row.get(lon_col, '') if lon_col else ''
+        try:
+            parts.append(str(round(float(lat), 5)) if pd.notna(lat) and lat != '' else '')
+        except Exception:
+            parts.append('')
+        try:
+            parts.append(str(round(float(lon), 5)) if pd.notna(lon) and lon != '' else '')
+        except Exception:
+            parts.append('')
+
+        # streets and ward
+        parts.append(_normalize_str(row.get(street1_col, '') if street1_col else ''))
+        parts.append(_normalize_str(row.get(street2_col, '') if street2_col else ''))
+        parts.append(_normalize_str(row.get(ward_col, '') if ward_col else ''))
+
+        # injuries: sum any injury-like columns
+        inj_sum = 0
+        for c in inj_cols:
+            try:
+                v = row.get(c, 0)
+                inj_sum += int(v) if pd.notna(v) and v != '' else 0
+            except Exception:
+                pass
+        parts.append(str(inj_sum))
+
+        # fatalities: sum any fatal-like columns
+        fat_sum = 0
+        for c in fat_cols:
+            try:
+                v = row.get(c, 0)
+                fat_sum += int(v) if pd.notna(v) and v != '' else 0
+            except Exception:
+                pass
+        parts.append(str(fat_sum))
+
+        # fallback uid
+        if uid:
+            parts.append(str(row.get(uid, '')))
+
+        s = '|'.join(parts)
+        h = hashlib.md5(s.encode('utf-8')).hexdigest()
+        val = int(h, 16) % n_buckets
+        return val
+
+    return df.apply(row_to_bucket, axis=1)
+
+
+def _add_date_features(df, date_col_candidates=None):
+    """Add simple date-derived numeric columns to the dataframe.
+
+    Adds: report_year, report_month, report_day, report_weekday, report_hour (where available).
+    If no date column is found, function is a no-op.
+    """
+    if date_col_candidates is None:
+        date_col_candidates = ['report_dat', 'reportdate', 'fromdate', 'lastupdatedate', 'date', 'occur_date']
+    colmap = {c.lower(): c for c in df.columns}
+    date_col = None
+    for cand in date_col_candidates:
+        if cand.lower() in colmap:
+            date_col = colmap[cand.lower()]
+            break
+    if date_col is None:
+        return
+    try:
+        ser = pd.to_datetime(df[date_col], errors='coerce')
+    except Exception:
+        ser = pd.to_datetime(df[date_col].astype(str), errors='coerce')
+
+    df['report_year'] = ser.dt.year.fillna(-1).astype(float)
+    df['report_month'] = ser.dt.month.fillna(-1).astype(float)
+    df['report_day'] = ser.dt.day.fillna(-1).astype(float)
+    df['report_weekday'] = ser.dt.weekday.fillna(-1).astype(float)
+    # hour may not exist; if parsing fails we'll get NaN
+    df['report_hour'] = ser.dt.hour.fillna(-1).astype(float)
+
+
+def _add_hashed_street(df, n_hash_buckets=32, street_col_candidates=None):
+    """Add a small hashed numeric feature for street/address text fields.
+
+    Adds `street_hash_0..N-1` as dense float columns containing one-hot-ish hashed values.
+    Uses MD5-based hashing reduced to a bucket and then maps to a small integer vector.
+    """
+    if street_col_candidates is None:
+        street_col_candidates = ['street1', 'street', 'address', 'mar_address', 'nearestintstreetname']
+    colmap = {c.lower(): c for c in df.columns}
+    street_col = None
+    for cand in street_col_candidates:
+        if cand.lower() in colmap:
+            street_col = colmap[cand.lower()]
+            break
+    if street_col is None:
+        return
+
+    import hashlib
+    # create a single integer hash bucket per row
+    def row_hash(val):
+        if pd.isna(val) or str(val).strip() == '':
+            return -1
+        h = hashlib.md5(str(val).encode('utf-8')).hexdigest()
+        return int(h, 16) % n_hash_buckets
+
+    buckets = df[street_col].apply(row_hash).fillna(-1).astype(int).to_numpy()
+    # create N numeric columns with a one-hot style (0/1) encoded as floats; missing bucket => zeros
+    for i in range(n_hash_buckets):
+        colname = f'street_hash_{i}'
+        df[colname] = (buckets == i).astype(float)
+
+
+def _add_latlon_bins(df, bins=20, lat_col_candidates=None, lon_col_candidates=None):
+    """Add coarse spatial bins for latitude/longitude and rounded lat/lon numeric features.
+
+    Adds: lat_round, lon_round, lat_bin, lon_bin (bins numbered 0..bins-1, -1 for missing).
+    """
+    if lat_col_candidates is None:
+        lat_col_candidates = ['latitude', 'mpdlatitude', 'lat']
+    if lon_col_candidates is None:
+        lon_col_candidates = ['longitude', 'mpdlongitude', 'lon']
+    colmap = {c.lower(): c for c in df.columns}
+    lat_col = None
+    lon_col = None
+    for cand in lat_col_candidates:
+        if cand.lower() in colmap:
+            lat_col = colmap[cand.lower()]
+            break
+    for cand in lon_col_candidates:
+        if cand.lower() in colmap:
+            lon_col = colmap[cand.lower()]
+            break
+    if lat_col is None or lon_col is None:
+        return
+    try:
+        lat = pd.to_numeric(df[lat_col], errors='coerce')
+        lon = pd.to_numeric(df[lon_col], errors='coerce')
+    except Exception:
+        lat = pd.to_numeric(df[lat_col].astype(str), errors='coerce')
+        lon = pd.to_numeric(df[lon_col].astype(str), errors='coerce')
+
+    df['lat_round'] = lat.round(3).fillna(0.0).astype(float)
+    df['lon_round'] = lon.round(3).fillna(0.0).astype(float)
+
+    try:
+        # compute bins using quantiles if possible to get balanced bins; fallback to linear bins
+        valid_lat = lat.dropna()
+        valid_lon = lon.dropna()
+        if len(valid_lat) >= bins and len(valid_lon) >= bins:
+            # qcut may produce NaNs for duplicates; use rank-based discretization
+            df['lat_bin'] = pd.qcut(lat.rank(method='first'), q=bins, labels=False, duplicates='drop')
+            df['lon_bin'] = pd.qcut(lon.rank(method='first'), q=bins, labels=False, duplicates='drop')
+        else:
+            lat_min, lat_max = valid_lat.min() if len(valid_lat) > 0 else 0.0, valid_lat.max() if len(valid_lat) > 0 else 0.0
+            lon_min, lon_max = valid_lon.min() if len(valid_lon) > 0 else 0.0, valid_lon.max() if len(valid_lon) > 0 else 0.0
+            lat_span = (lat_max - lat_min) + 1e-6
+            lon_span = (lon_max - lon_min) + 1e-6
+            df['lat_bin'] = (((lat - lat_min) / lat_span) * bins).fillna(-1).astype(int).clip(lower=-1, upper=bins-1)
+            df['lon_bin'] = (((lon - lon_min) / lon_span) * bins).fillna(-1).astype(int).clip(lower=-1, upper=bins-1)
+    except Exception:
+        # fallback: set -1 for bins
+        df['lat_bin'] = -1
+        df['lon_bin'] = -1
+
+
+# Debugging code - to be removed or commented out in production
+# python - <<'PY'
+# import pandas as pd
+# from data import generate_labels_for_df
+# df = pd.read_csv('data.csv', nrows=50, low_memory=False)
+# labs = generate_labels_for_df(df, n_buckets=100)
+# print(df[['REPORTDATE','LATITUDE','LONGITUDE','ADDRESS','WARD']].head().to_string())
+# print('labels:', list(labs[:20]))
+# PY
+
+# Command to run the training (to be executed in the terminal, not in the script)
+# python train.py data.csv --model-type mlp --generate-labels --n-buckets 100 --epochs 5 --batch-size 256 --lr 1e-3
+
diff --git a/roadcast/debug_labels.py b/roadcast/debug_labels.py
new file mode 100644
index 0000000..7529efd
--- /dev/null
+++ b/roadcast/debug_labels.py
@@ -0,0 +1,76 @@
+import pandas as pd
+import hashlib
+from data import _normalize_str, _normalize_date
+
+p='data.csv'
+df=pd.read_csv(p, nrows=50, low_memory=False)
+print('Columns:', list(df.columns))
+
+colmap = {c.lower(): c for c in df.columns}
+
+def get_col(*candidates):
+    for cand in candidates:
+        key = cand.lower()
+        if key in colmap:
+            return colmap[key]
+    return None
+
+report_col = get_col('report_dat', 'reportdate', 'fromdate', 'lastupdatedate')
+lat_col = get_col('latitude', 'mpdlatitude', 'lat')
+lon_col = get_col('longitude', 'mpdlongitude', 'lon')
+street1_col = get_col('street1', 'address', 'mar_address', 'nearestintstreetname')
+street2_col = get_col('street2', 'nearestintstreetname')
+ward_col = get_col('ward')
+inj_cols = [c for c in df.columns if 'INJUR' in c.upper()]
+fat_cols = [c for c in df.columns if 'FATAL' in c.upper()]
+uid = get_col('crimeid', 'eventid', 'objectid', 'ccn')
+
+print('Resolved columns:')
+print('report_col=', report_col)
+print('lat_col=', lat_col)
+print('lon_col=', lon_col)
+print('street1_col=', street1_col)
+print('street2_col=', street2_col)
+print('ward_col=', ward_col)
+print('inj_cols=', inj_cols[:10])
+print('fat_cols=', fat_cols[:10])
+print('uid=', uid)
+
+for i, row in df.iterrows():
+    parts = []
+    parts.append(_normalize_date(row.get(report_col, '') if report_col else ''))
+    lat = row.get(lat_col, '') if lat_col else ''
+    lon = row.get(lon_col, '') if lon_col else ''
+    try:
+        parts.append(str(round(float(lat), 5)) if pd.notna(lat) and lat != '' else '')
+    except Exception:
+        parts.append('')
+    try:
+        parts.append(str(round(float(lon), 5)) if pd.notna(lon) and lon != '' else '')
+    except Exception:
+        parts.append('')
+    parts.append(_normalize_str(row.get(street1_col, '') if street1_col else ''))
+    parts.append(_normalize_str(row.get(street2_col, '') if street2_col else ''))
+    parts.append(_normalize_str(row.get(ward_col, '') if ward_col else ''))
+    inj_sum = 0
+    for c in inj_cols:
+        try:
+            v = row.get(c, 0)
+            inj_sum += int(v) if pd.notna(v) and v != '' else 0
+        except Exception:
+            pass
+    parts.append(str(inj_sum))
+    fat_sum = 0
+    for c in fat_cols:
+        try:
+            v = row.get(c, 0)
+            fat_sum += int(v) if pd.notna(v) and v != '' else 0
+        except Exception:
+            pass
+    parts.append(str(fat_sum))
+    if uid:
+        parts.append(str(row.get(uid, '')))
+    s='|'.join(parts)
+    h=hashlib.md5(s.encode('utf-8')).hexdigest()
+    val=int(h,16)%100
+    print(i, 'label=', val, 's="'+s+'"')
diff --git a/roadcast/diagnostics.py b/roadcast/diagnostics.py
new file mode 100644
index 0000000..2bc7219
--- /dev/null
+++ b/roadcast/diagnostics.py
@@ -0,0 +1,67 @@
+import traceback
+import numpy as np
+import torch
+from torch.utils.data import DataLoader
+
+from data import CSVDataset
+from models import create_model
+
+
+def print_arr_stats(name, arr):
+    arr = np.asarray(arr)
+    print(f"{name}: shape={arr.shape} dtype={arr.dtype}")
+    print(f"  mean={np.nanmean(arr):.6f} std={np.nanstd(arr):.6f} min={np.nanmin(arr):.6f} max={np.nanmax(arr):.6f}")
+    print(f"  any_nan={np.isnan(arr).any()} any_inf={np.isinf(arr).any()}")
+
+
+def main():
+    try:
+        print('Loading dataset...')
+        ds = CSVDataset('data.csv', generate_labels=True, n_buckets=100)
+        print(f'dataset length: {len(ds)}')
+
+        X = ds.features.numpy()
+        y = ds.labels.numpy()
+
+        print_arr_stats('X (all)', X)
+        # per-column NaN/inf counts
+        nan_counts = np.isnan(X).sum(axis=0)
+        inf_counts = np.isinf(X).sum(axis=0)
+        print('per-column nan counts (first 20):', nan_counts[:20].tolist())
+        print('per-column inf counts (first 20):', inf_counts[:20].tolist())
+
+        print('Labels stats: unique count=', len(np.unique(y)))
+        print('Labels min/max:', int(y.min()), int(y.max()))
+        vals, counts = np.unique(y, return_counts=True)
+        print('Label distribution (first 20):', list(zip(vals[:20].tolist(), counts[:20].tolist())))
+
+        # get a small batch
+        dl = DataLoader(ds, batch_size=64, shuffle=False)
+        xb, yb = next(iter(dl))
+        print_arr_stats('xb batch', xb.numpy())
+        print('yb batch unique:', np.unique(yb.numpy()))
+
+        # build model
+        print('Building model...')
+        model = create_model(device='cpu', model_type='mlp', input_dim=X.shape[1], num_classes=100)
+        model.eval()
+        with torch.no_grad():
+            out = model(xb)
+        out_np = out.numpy()
+        print_arr_stats('model outputs', out_np)
+
+        # check for rows with NaN/Inf in features
+        bad_rows = np.where(np.isnan(X).any(axis=1) | np.isinf(X).any(axis=1))[0]
+        print('bad rows count:', len(bad_rows))
+        if len(bad_rows) > 0:
+            print('first bad row index:', bad_rows[0])
+            print('row values:', X[bad_rows[0]].tolist())
+            print('label:', int(y[bad_rows[0]]))
+
+    except Exception as e:
+        print('Exception during diagnostics:')
+        traceback.print_exc()
+
+
+if __name__ == '__main__':
+    main()
diff --git a/roadcast/fit_kmeans.py b/roadcast/fit_kmeans.py
new file mode 100644
index 0000000..0a43828
--- /dev/null
+++ b/roadcast/fit_kmeans.py
@@ -0,0 +1,65 @@
+"""Fit k-means centers on CSV numeric features (optionally PCA) and save centers to .npz
+
+Usage: python fit_kmeans.py data.csv --n-buckets 10 --out kmeans_centers_final.npz --sample 50000 --pca 50
+"""
+import argparse
+import numpy as np
+import pandas as pd
+
+from data import generate_kmeans_labels
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('csv')
+    parser.add_argument('--n-buckets', type=int, default=10)
+    parser.add_argument('--out', default='kmeans_centers_final.npz')
+    parser.add_argument('--sample', type=int, default=50000, help='max rows to sample for fitting')
+    parser.add_argument('--pca', type=int, default=0, help='Apply PCA to reduce dims before kmeans (0=none)')
+    args = parser.parse_args()
+
+    # read numeric columns only to avoid huge memory usage
+    df = pd.read_csv(args.csv, low_memory=False)
+    num_df = df.select_dtypes(include=['number']).fillna(0.0)
+    data = num_df.values.astype(float)
+    if data.shape[0] == 0 or data.shape[1] == 0:
+        raise SystemExit('No numeric data found in CSV')
+
+    # sample rows if requested
+    if args.sample and args.sample < data.shape[0]:
+        rng = np.random.default_rng(42)
+        idx = rng.choice(data.shape[0], size=args.sample, replace=False)
+        sample_data = data[idx]
+    else:
+        sample_data = data
+
+    # Use the kmeans implementation via generate_kmeans_labels for fitting centers.
+    # We'll call the internal function by adapting it here: import numpy locally.
+    import numpy as _np
+
+    # initialize centers by random sampling
+    rng = _np.random.default_rng(42)
+    k = min(args.n_buckets, sample_data.shape[0])
+    centers_idx = rng.choice(sample_data.shape[0], size=k, replace=False)
+    centers = sample_data[centers_idx].astype(float)
+
+    max_iters = 50
+    for _ in range(max_iters):
+        dists = np.linalg.norm(sample_data[:, None, :] - centers[None, :, :], axis=2)
+        labels = np.argmin(dists, axis=1)
+        new_centers = np.zeros_like(centers)
+        counts = np.zeros((centers.shape[0],), dtype=int)
+        for i, lab in enumerate(labels):
+            new_centers[lab] += sample_data[i]
+            counts[lab] += 1
+        for kk in range(centers.shape[0]):
+            if counts[kk] > 0:
+                new_centers[kk] = new_centers[kk] / counts[kk]
+            else:
+                new_centers[kk] = sample_data[rng.integers(0, sample_data.shape[0])]
+        shift = np.linalg.norm(new_centers - centers, axis=1).max()
+        centers = new_centers
+        if shift < 1e-4:
+            break
+
+    np.savez_compressed(args.out, centers=centers)
+    print('Saved centers to', args.out)
diff --git a/roadcast/inference.py b/roadcast/inference.py
new file mode 100644
index 0000000..49e8b6e
--- /dev/null
+++ b/roadcast/inference.py
@@ -0,0 +1,30 @@
+import os
+import torch
+import torch.nn.functional as F
+from PIL import Image
+from torchvision import transforms
+
+from models import create_model
+
+
+def load_model(path, device=None, in_channels=3, num_classes=10):
+    device = device or ('cuda' if torch.cuda.is_available() else 'cpu')
+    checkpoint = torch.load(path, map_location=device)
+    model = create_model(device=device, in_channels=in_channels, num_classes=num_classes)
+    model.load_state_dict(checkpoint['model_state_dict'])
+    model.eval()
+    class_to_idx = checkpoint.get('class_to_idx')
+    idx_to_class = {v: k for k, v in class_to_idx.items()} if class_to_idx else None
+    return model, idx_to_class
+
+
+def predict_image(model, img_path, device=None):
+    device = device or ('cuda' if torch.cuda.is_available() else 'cpu')
+    preprocess = transforms.Compose([transforms.Resize((224, 224)), transforms.ToTensor()])
+    img = Image.open(img_path).convert('RGB')
+    x = preprocess(img).unsqueeze(0).to(device)
+    with torch.no_grad():
+        logits = model(x)
+        probs = F.softmax(logits, dim=1)
+        conf, idx = torch.max(probs, dim=1)
+    return int(idx.item()), float(conf.item())
diff --git a/roadcast/inspect_csv.py b/roadcast/inspect_csv.py
new file mode 100644
index 0000000..4e2c0b2
--- /dev/null
+++ b/roadcast/inspect_csv.py
@@ -0,0 +1,34 @@
+import pandas as pd
+
+p = 'data.csv'
+print('Reading first 2000 rows of', p)
+df = pd.read_csv(p, nrows=2000, low_memory=False)
+print('Columns:', list(df.columns))
+cols = ['report_dat','latitude','longitude','street1','street2','ward','injuries','fatalities']
+print('\nField stats for label-generator columns:')
+for c in cols:
+    if c in df.columns:
+        ser = df[c]
+        try:
+            unique = ser.dropna().unique()[:5].tolist()
+        except Exception:
+            unique = []
+        print(f"{c}: present dtype={ser.dtype} n_unique={ser.nunique(dropna=False)} n_null={int(ser.isna().sum())} sample_values={unique}")
+    else:
+        print(f"{c}: MISSING")
+
+# If labels already present, show distribution
+if 'label' in df.columns:
+    print('\nLabel column present in sample:')
+    print(df['label'].value_counts().head(20))
+else:
+    print('\nLabel column not present in sample')
+
+# Also show per-column fraction NaN for numeric columns
+num_cols = df.select_dtypes(include=['number']).columns.tolist()
+print('\nNumeric columns and NaN fraction (first 20):')
+for c in num_cols[:20]:
+    ser = df[c]
+    print(f"{c}: n={len(ser)} null_frac={ser.isna().mean():.4f} min={ser.min()} max={ser.max()}")
+
+print('\nDone')
diff --git a/roadcast/kmeans_centers.npz b/roadcast/kmeans_centers.npz
new file mode 100644
index 0000000..5784c04
Binary files /dev/null and b/roadcast/kmeans_centers.npz differ
diff --git a/roadcast/kmeans_centers_all.npz b/roadcast/kmeans_centers_all.npz
new file mode 100644
index 0000000..867fba0
Binary files /dev/null and b/roadcast/kmeans_centers_all.npz differ
diff --git a/roadcast/kmeans_centers_best.npz b/roadcast/kmeans_centers_best.npz
new file mode 100644
index 0000000..2fdd1ce
Binary files /dev/null and b/roadcast/kmeans_centers_best.npz differ
diff --git a/roadcast/kmeans_centers_final.npz b/roadcast/kmeans_centers_final.npz
new file mode 100644
index 0000000..8859f5d
Binary files /dev/null and b/roadcast/kmeans_centers_final.npz differ
diff --git a/roadcast/kmeans_centers_nb10.npz b/roadcast/kmeans_centers_nb10.npz
new file mode 100644
index 0000000..2fdd1ce
Binary files /dev/null and b/roadcast/kmeans_centers_nb10.npz differ
diff --git a/roadcast/model.pth b/roadcast/model.pth
new file mode 100644
index 0000000..1cdfe6e
Binary files /dev/null and b/roadcast/model.pth differ
diff --git a/roadcast/models.py b/roadcast/models.py
new file mode 100644
index 0000000..a3e7aee
--- /dev/null
+++ b/roadcast/models.py
@@ -0,0 +1,68 @@
+import torch
+import torch.nn as nn
+
+
+class SimpleCNN(nn.Module):
+    """A small CNN for image classification (adjustable). Automatically computes flattened size."""
+    def __init__(self, in_channels=3, num_classes=10, input_size=(3, 224, 224)):
+        super().__init__()
+        self.features = nn.Sequential(
+            nn.Conv2d(in_channels, 32, kernel_size=3, padding=1),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+            nn.Conv2d(32, 64, kernel_size=3, padding=1),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+        )
+        # compute flatten size using a dummy tensor
+        with torch.no_grad():
+            dummy = torch.zeros(1, *input_size)
+            feat = self.features(dummy)
+            flat_features = int(feat.numel() / feat.shape[0])
+
+        self.classifier = nn.Sequential(
+            nn.Flatten(),
+            nn.Linear(flat_features, 256),
+            nn.ReLU(),
+            nn.Dropout(0.5),
+            nn.Linear(256, num_classes),
+        )
+
+    def forward(self, x):
+        x = self.features(x)
+        x = self.classifier(x)
+        return x
+
+
+class MLP(nn.Module):
+    """Simple MLP for tabular CSV data classification."""
+    def __init__(self, input_dim, hidden_dims=(256, 128), num_classes=2):
+        super().__init__()
+        layers = []
+        prev = input_dim
+        for h in hidden_dims:
+            layers.append(nn.Linear(prev, h))
+            layers.append(nn.ReLU())
+            layers.append(nn.Dropout(0.2))
+            prev = h
+        layers.append(nn.Linear(prev, num_classes))
+        self.net = nn.Sequential(*layers)
+
+    def forward(self, x):
+        return self.net(x)
+
+
+def create_model(device=None, in_channels=3, num_classes=10, input_size=(3, 224, 224), model_type='cnn', input_dim=None, hidden_dims=None):
+    if model_type == 'mlp':
+        if input_dim is None:
+            raise ValueError('input_dim is required for mlp model_type')
+        if hidden_dims is None:
+            model = MLP(input_dim=input_dim, num_classes=num_classes)
+        else:
+            model = MLP(input_dim=input_dim, hidden_dims=hidden_dims, num_classes=num_classes)
+    else:
+        model = SimpleCNN(in_channels=in_channels, num_classes=num_classes, input_size=input_size)
+
+    if device:
+        model.to(device)
+    return model
diff --git a/roadcast/openweather_inference.py b/roadcast/openweather_inference.py
new file mode 100644
index 0000000..3775112
--- /dev/null
+++ b/roadcast/openweather_inference.py
@@ -0,0 +1,339 @@
+"""
+Fetch OpenWeather data for a coordinate/time and run the trained MLP to predict the k-means cluster label.
+
+Usage examples:
+  # with training CSV provided to compute preprocessing stats:
+  python openweather_inference.py --lat 38.9 --lon -77.0 --datetime "2025-09-27T12:00:00" --train-csv data.csv --model model.pth --centers kmeans_centers_all.npz --api-key $OPENWEATHER_KEY
+
+  # with precomputed preprocess meta (saved from training):
+  python openweather_inference.py --lat 38.9 --lon -77.0 --datetime "2025-09-27T12:00:00" --preprocess-meta preprocess_meta.npz --model model.pth --centers kmeans_centers_all.npz --api-key $OPENWEATHER_KEY
+
+Notes:
+- The script uses the same feature-engineering helpers in `data.py` so the model sees identical inputs.
+- You must either provide `--train-csv` (to compute feature columns & means/stds) or `--preprocess-meta` previously saved.
+- Provide the OpenWeather API key via --api-key or the OPENWEATHER_KEY environment variable.
+"""
+
+import os
+import argparse
+import json
+from datetime import datetime
+import numpy as np
+import pandas as pd
+import torch
+import torch.nn.functional as F
+
+# reuse helpers from your repo
+from data import _add_date_features, _add_latlon_bins, _add_hashed_street, CSVDataset
+from inference import load_model
+
+# module-level caches to avoid reloading heavy artifacts per request
+_CACHED_MODEL = None
+_CACHED_IDX_TO_CLASS = None
+_CACHED_CENTERS = None
+_CACHED_PREPROCESS_META = None
+
+
+OW_BASE = 'https://api.openweathermap.org/data/2.5/onecall'
+
+
+def fetch_openweather(lat, lon, api_key, dt_iso=None):
+    """Fetch weather from OpenWeather One Call API for given lat/lon. If dt_iso provided, we fetch current+hourly and pick closest timestamp."""
+    try:
+        import requests
+    except Exception:
+        raise RuntimeError('requests library is required to fetch OpenWeather data')
+    params = {
+        'lat': float(lat),
+        'lon': float(lon),
+        'appid': api_key,
+        'units': 'metric',
+        'exclude': 'minutely,alerts'
+    }
+    r = requests.get(OW_BASE, params=params, timeout=10)
+    r.raise_for_status()
+    payload = r.json()
+    # if dt_iso provided, find nearest hourly data point
+    if dt_iso:
+        try:
+            target = pd.to_datetime(dt_iso)
+        except Exception:
+            target = None
+        best = None
+        if 'hourly' in payload and target is not None:
+            hours = payload['hourly']
+            best = min(hours, key=lambda h: abs(pd.to_datetime(h['dt'], unit='s') - target))
+            # convert keys to a flat dict with prefix 'ow_'
+            d = {
+                'ow_temp': best.get('temp'),
+                'ow_feels_like': best.get('feels_like'),
+                'ow_pressure': best.get('pressure'),
+                'ow_humidity': best.get('humidity'),
+                'ow_wind_speed': best.get('wind_speed'),
+                'ow_clouds': best.get('clouds'),
+                'ow_pop': best.get('pop'),
+            }
+            return d
+    # fallback: use current
+    cur = payload.get('current', {})
+    d = {
+        'ow_temp': cur.get('temp'),
+        'ow_feels_like': cur.get('feels_like'),
+        'ow_pressure': cur.get('pressure'),
+        'ow_humidity': cur.get('humidity'),
+        'ow_wind_speed': cur.get('wind_speed'),
+        'ow_clouds': cur.get('clouds'),
+        'ow_pop': None,
+    }
+    return d
+
+
+def fetch_roadrisk(roadrisk_url, api_key=None):
+    """Fetch the RoadRisk endpoint (expects JSON). If `api_key` is provided, we'll attach it as a query param if the URL has no key.
+
+    We flatten top-level numeric fields into `rr_*` keys for the feature row.
+    """
+    # if api_key provided and url does not contain appid, append it
+    try:
+        import requests
+    except Exception:
+        raise RuntimeError('requests library is required to fetch RoadRisk data')
+    url = roadrisk_url
+    if api_key and 'appid=' not in roadrisk_url:
+        sep = '&' if '?' in roadrisk_url else '?'
+        url = f"{roadrisk_url}{sep}appid={api_key}"
+
+    r = requests.get(url, timeout=10)
+    r.raise_for_status()
+    payload = r.json()
+    # flatten numeric top-level fields
+    out = {}
+    if isinstance(payload, dict):
+        for k, v in payload.items():
+            if isinstance(v, (int, float)):
+                out[f'rr_{k}'] = v
+            # if nested objects contain simple numeric fields, pull them too (one level deep)
+            elif isinstance(v, dict):
+                for kk, vv in v.items():
+                    if isinstance(vv, (int, float)):
+                        out[f'rr_{k}_{kk}'] = vv
+    return out
+
+
+def build_row(lat, lon, dt_iso=None, street=None, extra_weather=None):
+    """Construct a single-row DataFrame with columns expected by the training pipeline.
+
+    It intentionally uses column names the original `data.py` looked for (REPORTDATE, LATITUDE, LONGITUDE, ADDRESS, etc.).
+    """
+    row = {}
+    # date column matching common names
+    row['REPORTDATE'] = dt_iso if dt_iso else datetime.utcnow().isoformat()
+    row['LATITUDE'] = lat
+    row['LONGITUDE'] = lon
+    row['ADDRESS'] = street if street else ''
+    # include some injury/fatality placeholders that the label generator expects
+    row['INJURIES'] = 0
+    row['FATALITIES'] = 0
+    # include weather features returned by OpenWeather (prefixed 'ow_')
+    if extra_weather:
+        for k, v in extra_weather.items():
+            row[k] = v
+    return pd.DataFrame([row])
+
+
+def prepare_features(df_row, train_csv=None, preprocess_meta=None, feature_engineer=True, lat_lon_bins=20):
+    """Given a one-row DataFrame, apply same feature engineering and standardization as training.
+
+    If preprocess_meta is provided (npz), use it. Otherwise train_csv must be provided to compute stats.
+    Returns a torch.FloatTensor of shape (1, input_dim) and the feature_columns list.
+    """
+    # apply feature engineering helpers
+    if feature_engineer:
+        try:
+            _add_date_features(df_row)
+        except Exception:
+            pass
+        try:
+            _add_latlon_bins(df_row, bins=lat_lon_bins)
+        except Exception:
+            pass
+        try:
+            _add_hashed_street(df_row)
+        except Exception:
+            pass
+
+    # if meta provided, load feature_columns, means, stds
+    if preprocess_meta and os.path.exists(preprocess_meta):
+        meta = np.load(preprocess_meta, allow_pickle=True)
+        feature_columns = meta['feature_columns'].tolist()
+        means = meta['means']
+        stds = meta['stds']
+    else:
+        if not train_csv:
+            raise ValueError('Either preprocess_meta or train_csv must be provided to derive feature stats')
+        # instantiate a CSVDataset on train_csv (feature_engineer True) to reuse its preprocessing
+        ds = CSVDataset(train_csv, feature_columns=None, label_column='label', generate_labels=False, n_buckets=10, label_method='kmeans', label_store=None, feature_engineer=feature_engineer, lat_lon_bins=lat_lon_bins, nrows=None)
+        feature_columns = ds.feature_columns
+        means = ds.feature_means
+        stds = ds.feature_stds
+        # save meta for reuse
+        np.savez_compressed('preprocess_meta.npz', feature_columns=np.array(feature_columns, dtype=object), means=means, stds=stds)
+        print('Saved preprocess_meta.npz')
+
+    # ensure all feature columns exist in df_row
+    for c in feature_columns:
+        if c not in df_row.columns:
+            df_row[c] = 0
+
+    # coerce and fill using means
+    features_df = df_row[feature_columns].apply(lambda c: pd.to_numeric(c, errors='coerce'))
+    features_df = features_df.fillna(pd.Series(means, index=feature_columns)).fillna(0.0)
+    # standardize
+    features_np = (features_df.values - means) / (stds + 1e-6)
+    import torch
+    return torch.tensor(features_np, dtype=torch.float32), feature_columns
+
+
+def predict_from_openweather(lat, lon, dt_iso=None, street=None, api_key=None, train_csv=None, preprocess_meta=None, model_path='model.pth', centers_path='kmeans_centers_all.npz', roadrisk_url=None):
+    api_key = api_key or os.environ.get('OPENWEATHER_KEY')
+    if api_key is None:
+        raise ValueError('OpenWeather API key required via --api-key or OPENWEATHER_KEY env var')
+
+    # gather weather/road-risk features
+    weather = {}
+    if roadrisk_url:
+        try:
+            rr = fetch_roadrisk(roadrisk_url, api_key=api_key)
+            weather.update(rr)
+        except Exception as e:
+            print('Warning: failed to fetch roadrisk URL:', e)
+    else:
+        try:
+            ow = fetch_openweather(lat, lon, api_key, dt_iso=dt_iso)
+            weather.update(ow)
+        except Exception as e:
+            print('Warning: failed to fetch openweather:', e)
+
+    df_row = build_row(lat, lon, dt_iso=dt_iso, street=street, extra_weather=weather)
+    x_tensor, feature_columns = prepare_features(df_row, train_csv=train_csv, preprocess_meta=preprocess_meta)
+
+    # load model (infer num_classes from centers file if possible)
+    global _CACHED_MODEL, _CACHED_IDX_TO_CLASS, _CACHED_CENTERS, _CACHED_PREPROCESS_META
+
+    # ensure we have preprocess_meta available (prefer supplied path, otherwise fallback to saved file)
+    if preprocess_meta is None:
+        candidate = os.path.join(os.getcwd(), 'preprocess_meta.npz')
+        if os.path.exists(candidate):
+            preprocess_meta = candidate
+
+    # load centers (cache across requests)
+    if _CACHED_CENTERS is None:
+        if centers_path and os.path.exists(centers_path):
+            try:
+                npz = np.load(centers_path)
+                _CACHED_CENTERS = npz['centers']
+            except Exception:
+                _CACHED_CENTERS = None
+        else:
+            _CACHED_CENTERS = None
+
+    num_classes = _CACHED_CENTERS.shape[0] if _CACHED_CENTERS is not None else 10
+
+    # load model once and cache it
+    if _CACHED_MODEL is None:
+        try:
+            _CACHED_MODEL, _CACHED_IDX_TO_CLASS = load_model(model_path, device=None, in_channels=3, num_classes=num_classes)
+            device = 'cuda' if torch.cuda.is_available() else 'cpu'
+            _CACHED_MODEL.to(device)
+        except Exception as e:
+            raise
+    model = _CACHED_MODEL
+    idx_to_class = _CACHED_IDX_TO_CLASS
+    device = 'cuda' if torch.cuda.is_available() else 'cpu'
+    x_tensor = x_tensor.to(device)
+    with torch.no_grad():
+        logits = model(x_tensor)
+        probs = F.softmax(logits, dim=1).cpu().numpy()[0]
+        pred_idx = int(probs.argmax())
+        confidence = float(probs.max())
+
+    # optionally provide cluster centroid info
+    centroid = _CACHED_CENTERS[pred_idx] if _CACHED_CENTERS is not None else None
+
+    return {
+        'pred_cluster': int(pred_idx),
+        'confidence': confidence,
+        'probabilities': probs.tolist(),
+        'centroid': centroid.tolist() if centroid is not None else None,
+        'feature_columns': feature_columns,
+        'used_preprocess_meta': preprocess_meta
+    }
+
+
+def init_inference(model_path='model.pth', centers_path='kmeans_centers_all.npz', preprocess_meta=None):
+    """Eagerly load model, centers, and preprocess_meta into module-level caches.
+
+    This is intended to be called at app startup to surface load errors early and avoid
+    per-request disk IO. The function is best-effort and will print warnings if artifacts
+    are missing.
+    """
+    global _CACHED_MODEL, _CACHED_IDX_TO_CLASS, _CACHED_CENTERS, _CACHED_PREPROCESS_META
+
+    # prefer existing saved preprocess_meta if not explicitly provided
+    if preprocess_meta is None:
+        candidate = os.path.join(os.getcwd(), 'preprocess_meta.npz')
+        if os.path.exists(candidate):
+            preprocess_meta = candidate
+
+    _CACHED_PREPROCESS_META = preprocess_meta
+
+    # load centers
+    if _CACHED_CENTERS is None:
+        if centers_path and os.path.exists(centers_path):
+            try:
+                npz = np.load(centers_path)
+                _CACHED_CENTERS = npz['centers']
+                print(f'Loaded centers from {centers_path}')
+            except Exception as e:
+                print('Warning: failed to load centers:', e)
+                _CACHED_CENTERS = None
+        else:
+            print('No centers file found at', centers_path)
+            _CACHED_CENTERS = None
+
+    num_classes = _CACHED_CENTERS.shape[0] if _CACHED_CENTERS is not None else 10
+
+    # load model
+    if _CACHED_MODEL is None:
+        try:
+            _CACHED_MODEL, _CACHED_IDX_TO_CLASS = load_model(model_path, device=None, in_channels=3, num_classes=num_classes)
+            device = 'cuda' if torch.cuda.is_available() else 'cpu'
+            _CACHED_MODEL.to(device)
+            print(f'Loaded model from {model_path}')
+        except Exception as e:
+            print('Warning: failed to load model:', e)
+            _CACHED_MODEL = None
+
+    return {
+        'model_loaded': _CACHED_MODEL is not None,
+        'centers_loaded': _CACHED_CENTERS is not None,
+        'preprocess_meta': _CACHED_PREPROCESS_META
+    }
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--lat', type=float, required=True)
+    parser.add_argument('--lon', type=float, required=True)
+    parser.add_argument('--datetime', default=None, help='ISO datetime string to query hourly weather (optional)')
+    parser.add_argument('--street', default='')
+    parser.add_argument('--api-key', default=None, help='OpenWeather API key or use OPENWEATHER_KEY env var')
+    parser.add_argument('--train-csv', default=None, help='Path to training CSV to compute preprocessing stats (optional if --preprocess-meta provided)')
+    parser.add_argument('--preprocess-meta', default=None, help='Path to precomputed preprocess_meta.npz (optional)')
+    parser.add_argument('--model', default='model.pth')
+    parser.add_argument('--centers', default='kmeans_centers_all.npz')
+    parser.add_argument('--roadrisk-url', default=None, help='Optional custom RoadRisk API URL (if provided, will be queried instead of OneCall)')
+    args = parser.parse_args()
+
+    out = predict_from_openweather(args.lat, args.lon, dt_iso=args.datetime, street=args.street, api_key=args.api_key, train_csv=args.train_csv, preprocess_meta=args.preprocess_meta, model_path=args.model, centers_path=args.centers, roadrisk_url=args.roadrisk_url)
+    print(json.dumps(out, indent=2))
diff --git a/roadcast/requirements.txt b/roadcast/requirements.txt
new file mode 100644
index 0000000..ee9c857
--- /dev/null
+++ b/roadcast/requirements.txt
@@ -0,0 +1,5 @@
+flask>=2.0
+torch>=1.13
+torchvision>=0.14
+Pillow>=9.0
+tqdm>=4.60
diff --git a/roadcast/run_batch_inference.py b/roadcast/run_batch_inference.py
new file mode 100644
index 0000000..119610d
--- /dev/null
+++ b/roadcast/run_batch_inference.py
@@ -0,0 +1,113 @@
+import os
+import argparse
+import pandas as pd
+import numpy as np
+import time
+
+import openweather_inference as owi
+
+
+def find_column(df_cols, candidates):
+    cmap = {c.lower(): c for c in df_cols}
+    for cand in candidates:
+        if cand.lower() in cmap:
+            return cmap[cand.lower()]
+    return None
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('csv', help='Path to data CSV (e.g., data.csv)')
+    parser.add_argument('--out', default='inference_results.csv')
+    parser.add_argument('--lat-col', default=None)
+    parser.add_argument('--lon-col', default=None)
+    parser.add_argument('--date-col', default=None)
+    parser.add_argument('--model', default='model.pth')
+    parser.add_argument('--centers', default='kmeans_centers_all.npz')
+    parser.add_argument('--preprocess-meta', default=None)
+    parser.add_argument('--api-key', default=None)
+    parser.add_argument('--live', action='store_true', help='If set, call external RoadRisk/OpenWeather per row')
+    parser.add_argument('--roadrisk-url', default=None, help='Optional per-request RoadRisk URL to use when --live')
+    parser.add_argument('--subset', type=int, default=0, help='Process only first N rows for testing')
+    args = parser.parse_args()
+
+    df = pd.read_csv(args.csv, low_memory=False)
+    nrows = args.subset if args.subset and args.subset > 0 else len(df)
+    df = df.iloc[:nrows].copy()
+
+    # find sensible columns
+    lat_col = args.lat_col or find_column(df.columns, ['latitude', 'lat', 'mpdlatitude'])
+    lon_col = args.lon_col or find_column(df.columns, ['longitude', 'lon', 'mpdlongitude'])
+    date_col = args.date_col or find_column(df.columns, ['report_dat', 'reportdate', 'fromdate', 'lastupdatedate', 'date', 'occur_date'])
+
+    if lat_col is None or lon_col is None:
+        raise SystemExit('Could not find latitude/longitude columns automatically. Pass --lat-col and --lon-col.')
+
+    print(f'Using lat column: {lat_col}, lon column: {lon_col}, date column: {date_col}')
+
+    # eager init caches
+    status = owi.init_inference(model_path=args.model, centers_path=args.centers, preprocess_meta=args.preprocess_meta)
+    print('init status:', status)
+
+    results = []
+    t0 = time.time()
+    for i, row in df.iterrows():
+        lat = row.get(lat_col)
+        lon = row.get(lon_col)
+        dt = row.get(date_col) if date_col else None
+
+        try:
+            if args.live:
+                # call the full pipeline which may hit remote API
+                out = owi.predict_from_openweather(lat, lon, dt_iso=dt, street=None, api_key=args.api_key, train_csv=None, preprocess_meta=args.preprocess_meta, model_path=args.model, centers_path=args.centers, roadrisk_url=args.roadrisk_url)
+            else:
+                # local-only path: build row, prepare features using preprocess_meta, and run cached model
+                df_row = owi.build_row(lat, lon, dt_iso=dt, street=None, extra_weather=None)
+                x_tensor, feature_columns = owi.prepare_features(df_row, train_csv=None, preprocess_meta=args.preprocess_meta)
+                # ensure model cached
+                if owi._CACHED_MODEL is None:
+                    owi.init_inference(model_path=args.model, centers_path=args.centers, preprocess_meta=args.preprocess_meta)
+                model = owi._CACHED_MODEL
+                centers = owi._CACHED_CENTERS
+                device = 'cuda' if __import__('torch').cuda.is_available() else 'cpu'
+                model.to(device)
+                xt = x_tensor.to(device)
+                import torch
+                import torch.nn.functional as F
+                with torch.no_grad():
+                    logits = model(xt)
+                    probs = F.softmax(logits, dim=1).cpu().numpy()[0]
+                    pred_idx = int(probs.argmax())
+                    confidence = float(probs.max())
+                out = {'pred_cluster': pred_idx, 'confidence': confidence, 'probabilities': probs.tolist(), 'centroid': centers[pred_idx].tolist() if centers is not None else None, 'feature_columns': feature_columns}
+        except Exception as e:
+            out = {'error': str(e)}
+
+        # combine row and output into flat result
+        res = {
+            'orig_index': i,
+            'lat': lat,
+            'lon': lon,
+            'datetime': str(dt),
+        }
+        if 'error' in out:
+            res.update({'error': out['error']})
+        else:
+            res.update({
+                'pred_cluster': int(out.get('pred_cluster')),
+                'confidence': float(out.get('confidence')),
+            })
+        results.append(res)
+
+        if (len(results) % 50) == 0:
+            print(f'Processed {len(results)}/{nrows} rows...')
+
+    elapsed = time.time() - t0
+    print(f'Finished {len(results)} rows in {elapsed:.2f}s')
+    out_df = pd.DataFrame(results)
+    out_df.to_csv(args.out, index=False)
+    print('Wrote', args.out)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/roadcast/tests/smoke_predict.py b/roadcast/tests/smoke_predict.py
new file mode 100644
index 0000000..b215b0d
--- /dev/null
+++ b/roadcast/tests/smoke_predict.py
@@ -0,0 +1,23 @@
+import sys
+import types
+import os
+
+# Ensure repo root on path
+sys.path.insert(0, r"C:\Users\Samarth Jain\Documents\roadcast")
+
+# Create a fake openweather_inference module with a predictable function
+mod = types.ModuleType("openweather_inference")
+
+def predict_from_openweather(lat, lon, dt_iso=None, street='', api_key=None, train_csv=None, preprocess_meta=None, model_path=None, centers_path=None, roadrisk_url=None):
+    return {"label": 5, "confidence": 0.87, "lat": lat, "lon": lon, "dt": dt_iso}
+
+mod.predict_from_openweather = predict_from_openweather
+sys.modules["openweather_inference"] = mod
+
+# Import the Flask app and use its test client
+from app import app
+
+c = app.test_client()
+res = c.post("/predict-roadrisk", json={"lat": 38.9, "lon": -77.0})
+print("STATUS:", res.status_code)
+print("JSON:", res.get_json())
diff --git a/roadcast/tmp_infer.csv b/roadcast/tmp_infer.csv
new file mode 100644
index 0000000..1af6b82
--- /dev/null
+++ b/roadcast/tmp_infer.csv
@@ -0,0 +1,11 @@
+orig_index,lat,lon,datetime,error
+0,38.91557,-77.031697,2011/03/06 05:00:00+00,Either preprocess_meta or train_csv must be provided to derive feature stats
+1,38.875558,-77.017556,2011/03/06 08:45:00+00,Either preprocess_meta or train_csv must be provided to derive feature stats
+2,38.872976,-77.016987,2011/03/05 05:00:00+00,Either preprocess_meta or train_csv must be provided to derive feature stats
+3,38.929433,-77.003943,2011/03/08 05:00:00+00,Either preprocess_meta or train_csv must be provided to derive feature stats
+4,38.89674,-77.027034,2011/03/08 17:18:00+00,Either preprocess_meta or train_csv must be provided to derive feature stats
+5,38.89093,-76.993494,2011/03/12 05:00:00+00,Either preprocess_meta or train_csv must be provided to derive feature stats
+6,38.908478,-77.040086,2011/03/12 05:00:00+00,Either preprocess_meta or train_csv must be provided to derive feature stats
+7,38.846563,-76.976504,2011/03/12 05:00:00+00,Either preprocess_meta or train_csv must be provided to derive feature stats
+8,38.894783,-77.01292,2011/03/12 18:30:00+00,Either preprocess_meta or train_csv must be provided to derive feature stats
+9,38.934204,-77.034567,2011/03/14 04:00:00+00,Either preprocess_meta or train_csv must be provided to derive feature stats
diff --git a/roadcast/train.py b/roadcast/train.py
new file mode 100644
index 0000000..c646093
--- /dev/null
+++ b/roadcast/train.py
@@ -0,0 +1,158 @@
+import os
+import time
+import torch
+from torch import nn, optim
+from torch.utils.data import DataLoader, random_split
+from tqdm import tqdm
+
+from data import ImageFolderDataset, CSVDataset
+from models import create_model
+
+
+def train(dataset_root, epochs=3, batch_size=16, lr=1e-3, device=None, num_classes=10, model_type='cnn', csv_label='label', generate_labels=False, n_buckets=100, label_method='md5', label_store=None, feature_engineer=False, lat_lon_bins=20, nrows=None, seed=42, hidden_dims=None, weight_decay=0.0):
+    device = device or ('cuda' if torch.cuda.is_available() else 'cpu')
+    # Detect CSV vs folder dataset
+    if os.path.isfile(dataset_root) and dataset_root.lower().endswith('.csv'):
+        dataset = CSVDataset(dataset_root,
+                             label_column=csv_label,
+                             generate_labels=generate_labels,
+                             n_buckets=n_buckets,
+                             label_method=label_method,
+                             label_store=label_store,
+                             feature_engineer=feature_engineer,
+                             lat_lon_bins=lat_lon_bins,
+                             nrows=nrows)
+        # seed numpy/torch RNGs for reproducibility in experiments
+        try:
+            import numpy as _np
+            _np.random.seed(seed)
+        except Exception:
+            pass
+        try:
+            import random as _py_random
+            _py_random.seed(seed)
+        except Exception:
+            pass
+        try:
+            import torch as _torch
+            _torch.manual_seed(seed)
+            if _torch.cuda.is_available():
+                _torch.cuda.manual_seed_all(seed)
+        except Exception:
+            pass
+        # determine input dim for MLP
+        input_dim = dataset.features.shape[1]
+        # persist preprocessing metadata so inference can reuse identical stats
+        try:
+            import numpy as _np
+            meta_path = os.path.join(os.getcwd(), 'preprocess_meta.npz')
+            _np.savez_compressed(meta_path, feature_columns=_np.array(dataset.feature_columns, dtype=object), means=dataset.feature_means, stds=dataset.feature_stds)
+            print(f'Saved preprocess meta to {meta_path}')
+        except Exception:
+            pass
+        if model_type == 'cnn':
+            raise ValueError('CSV dataset should use model_type="mlp"')
+        # if we generated labels, infer the actual number of classes from the dataset labels
+        if generate_labels and hasattr(dataset, 'labels'):
+            try:
+                model_num_classes = int(dataset.labels.max().item()) + 1
+            except Exception:
+                model_num_classes = n_buckets
+        else:
+            model_num_classes = n_buckets if generate_labels else num_classes
+        # parse hidden_dims if provided by caller (tuple or list)
+        model = create_model(device=device, model_type='mlp', input_dim=input_dim, num_classes=model_num_classes, hidden_dims=hidden_dims)
+    else:
+        # assume folder of images
+        dataset = ImageFolderDataset(dataset_root)
+        model = create_model(device=device, model_type='cnn', input_size=(3, 224, 224), num_classes=num_classes)
+
+    # simple train/val split
+    val_size = max(1, int(0.1 * len(dataset)))
+    train_size = len(dataset) - val_size
+    train_set, val_set = random_split(dataset, [train_size, val_size])
+    train_loader = DataLoader(train_set, batch_size=batch_size, shuffle=True)
+    val_loader = DataLoader(val_set, batch_size=batch_size, shuffle=False)
+
+    criterion = nn.CrossEntropyLoss()
+    optimizer = optim.Adam(model.parameters(), lr=lr, weight_decay=weight_decay)
+
+    best_val_acc = 0.0
+    best_path = None
+
+    for epoch in range(epochs):
+        model.train()
+        running_loss = 0.0
+        pbar = tqdm(train_loader, desc=f"Epoch {epoch+1}/{epochs}")
+        for xb, yb in pbar:
+            xb = xb.to(device)
+            yb = yb.to(device)
+            optimizer.zero_grad()
+            outputs = model(xb)
+            loss = criterion(outputs, yb)
+            loss.backward()
+            optimizer.step()
+            running_loss += loss.item()
+            pbar.set_postfix(loss=running_loss / (pbar.n + 1))
+
+        # validation
+        model.eval()
+        correct = 0
+        total = 0
+        with torch.no_grad():
+            for xb, yb in val_loader:
+                xb = xb.to(device)
+                yb = yb.to(device)
+                outputs = model(xb)
+                preds = outputs.argmax(dim=1)
+                correct += (preds == yb).sum().item()
+                total += yb.size(0)
+        val_acc = correct / total if total > 0 else 0.0
+        print(f"Epoch {epoch+1} val_acc={val_acc:.4f}")
+
+        # save best
+        if val_acc > best_val_acc:
+            out_path = os.path.join(os.getcwd(), 'model.pth')
+            if hasattr(dataset, 'class_to_idx'):
+                meta = {'model_state_dict': model.state_dict(), 'class_to_idx': dataset.class_to_idx}
+            else:
+                meta = {'model_state_dict': model.state_dict()}
+            torch.save(meta, out_path)
+            best_val_acc = val_acc
+            best_path = out_path
+            print(f"Saved best model to {out_path} (val_acc={val_acc:.4f})")
+
+    return best_path
+
+
+if __name__ == '__main__':
+    import argparse
+    parser = argparse.ArgumentParser()
+    parser.add_argument('data_root')
+    parser.add_argument('--epochs', type=int, default=3)
+    parser.add_argument('--batch-size', type=int, default=16)
+    parser.add_argument('--lr', type=float, default=1e-3)
+    parser.add_argument('--model-type', choices=['cnn', 'mlp'], default='cnn')
+    parser.add_argument('--csv-label', default='label')
+    parser.add_argument('--generate-labels', action='store_true', help='If set, generate labels from columns instead of expecting label column')
+    parser.add_argument('--n-buckets', type=int, default=100, help='Number of label buckets when generating labels')
+    parser.add_argument('--label-method', choices=['md5', 'kmeans'], default='md5', help='Method to generate labels when --generate-labels is set')
+    parser.add_argument('--label-store', default=None, help='Path to save/load label metadata (e.g., kmeans centers .npz)')
+    parser.add_argument('--subset', type=int, default=0, help='If set (>0), load only first N rows from CSV for fast experiments')
+    parser.add_argument('--feature-engineer', action='store_true', help='If set, add simple date and lat/lon engineered features')
+    parser.add_argument('--lat-lon-bins', type=int, default=20, help='Number of bins for lat/lon coarse spatial features')
+    parser.add_argument('--seed', type=int, default=42, help='Random seed for experiments')
+    parser.add_argument('--hidden-dims', type=str, default='', help='Comma-separated hidden dims for MLP, e.g. "256,128"')
+    parser.add_argument('--weight-decay', type=float, default=0.0, help='Weight decay (L2) for optimizer')
+    args = parser.parse_args()
+    data_root = args.data_root
+    nrows = args.subset if args.subset > 0 else None
+    # parse hidden dims
+    hidden_dims = None
+    if args.hidden_dims:
+        try:
+            hidden_dims = tuple(int(x) for x in args.hidden_dims.split(',') if x.strip())
+        except Exception:
+            hidden_dims = None
+    train(data_root, epochs=args.epochs, batch_size=args.batch_size, lr=args.lr, model_type=args.model_type, csv_label=args.csv_label, generate_labels=args.generate_labels, n_buckets=args.n_buckets, label_method=args.label_method, label_store=args.label_store, feature_engineer=args.feature_engineer, lat_lon_bins=args.lat_lon_bins, nrows=nrows, seed=args.seed, hidden_dims=hidden_dims, weight_decay=args.weight_decay)
+