Decode SIH 2026-Bangalore

Pravah

Pravah is a digital twin of a Bengaluru arterial that answers counterfactual questions about traffic: not what traffic is, but what traffic would be if you changed something. Describe a scenario in plain English, such as "close a lane near Marathahalli" or "what if 50 percent more traffic arrived during peak", and the Gemini API turns it into a validated simulation specification. It picks from a numbered list of real corridor links rather than generating an identifier, so it structurally cannot invent a location, and it shows a plain-English readback of what it understood before anything runs. The system then executes two SUMO microsimulations on identical random seeds, one with the change and one without. The measured difference is the answer, rendered on a map with the cause marked separately from the downstream effect. We model Old Airport Road, Domlur to Marathahalli: twelve signalised junctions, calibrated for Indian mixed traffic using SUMO's sublane model so two-wheelers filter between lanes rather than queuing single file. The corridor was chosen by a survey script that ranked five candidates on how much room signal control actually has. Tavily reads live Bengaluru traffic news and matches incidents to specific corridor links, stating its confidence and refusing to guess when it cannot place one. An n8n workflow runs that on a schedule and delivers a daily digest. Every result is at least five simulation runs on different seeds with the spread reported, and every number carries its seeds, simulator version and git commit on screen. The rule that a gridlocked run counts as a failure rather than a fast run was written before any results existed.