Rain vs clear
Predict 09:00 rain, note Technopark speed. Switch to clear, predict again. Speeds should rise — same places, different scenario.
Compiling the hour
Compiling the hour
Prediction studio
Set date, time, and scenario for Thiruvananthapuram. See predicted speeds at named places, then choose start and end for a PPO corridor recommendation.
Hybrid GAT-LSTM
Speeds are predicted per named place in Thiruvananthapuram, not anonymous road indices.
PPO
Choose a start place and an end place. PPO picks the fastest predicted corridor between them.
Start place
Not selected yet
End place
Not selected yet
Type a place name. Results come from OpenStreetMap (Nominatim + Overpass), debounced while you type.
How to use this
Hybrid forecast — pick date, time, and scenario (normal, rain, event, accident, heavy, clear). Press Predict. You get ranked speeds for places across Thiruvananthapuram: Palayam, Kowdiar, Technopark, Medical College, and more.
PPO route — search start and end places (OSM-backed list with debounce). The agent recommends the best corridor between them using the latest predicted speeds, and names the places in plain language — not only raw road IDs.
Normal is the baseline. Rain and heavy traffic lower speeds. Accident drops them further around the graph. Clear raises them. The same inputs always produce the same demo forecast, so demos and reports stay repeatable.
This running app uses a lightweight demo predictor keyed by date, time, and scenario — not live city sensors. It is built to feel like peak-hour and weather effects on the real OSM graph. The full GAT–LSTM stack exists in the project for heavier lab runs; this page is the interactive face of that idea.
Place cards show km/h at the nearest snapped road. Green-ish values are freer flow; lower numbers are congested. After PPO, the note describes the recommended corridor between your two places and the predicted speed along that recommendation.
Open Map and generate the same start–end under the same scenario to see the speeds drawn on the city. Open Influence and load a heat map near one of your places to see which nearby roads are strongly linked in the model. Detect can add a camera count when you have a photo.
Speeds are not live Google traffic. Places come from OSM plus a curated list. YOLO is separate and only runs when you upload an image. Use this studio to think and demo; treat numbers as a consistent simulation on a real capital-city graph.
Sample workflows
Predict 09:00 rain, note Technopark speed. Switch to clear, predict again. Speeds should rise — same places, different scenario.
PPO from East Fort to Secretariat at 10:00 normal. Copy the place names into Map and generate for the same pair.
Predict heavy traffic, then PPO Kazhakkoottam → Pattom. Check the recommended corridor note uses real place names.
Forecast near Medical College / Ulloor. Lower evening speeds show rush-hour logic in the demo predictor.
Airport → Palayam under accident scenario. Expect harsher speeds; then open Influence at Airport to see local road links.
Run the same date/time/scenario twice. Ranked place speeds should match — that proves the demo is deterministic.
Common questions
Rain, accident, and heavy traffic lower the base speed field on purpose. Clear raises it. That is how you demo policy without live sensors.
A start place and an end place. It uses the latest prediction window for that timestamp and scenario when available.
Yes — type at least a couple of characters. Results come from Nominatim plus a cached Overpass catalog for Thiruvananthapuram.
Treat on-screen series as a studio instrument. Published MAE/R² live on Research for the trained model families.