FastAndSlow

Autonomous Wildfire Drone power byt the brand new jev model from typesafeai for quick reflexes and devin for deep reasoning tasks

Sobre el proyecto

FastAndSlow is an autonomous wildfire incident-response sandbox built around two complementary agents. Jev reacts quickly, scoring time-local hotspot evidence, opening simulated incidents, and assigning verification drones. Devin reasons more slowly about complex response plans. Every proposed route or action passes through a deterministic policy engine that can approve it, reject unsafe paths, or request a revision. The interface visualizes historical Deepfire satellite data, observed heat, nearby infrastructure, drone routes, uncertainty, and a complete audit trail.

Operators can also join a real-time Vonage Video API briefing with Ari, a simulated field liaison. They ask questions by voice; SLNG transcribes and synthesizes speech, while a conversational model receives a fresh read-only snapshot of the replay and recent conversation history. This lets Ari explain why a drone was rerouted, what assets may be at risk, and where evidence is insufficient without repeating canned answers.