The Fire Monitor system, developed by the Strela team during the 2026 Space Hackathon, utilizes a GeoAI agent to automate forest fire monitoring. The platform integrates satellite data from VIIRS and MODIS, processes burn severity using Sentinel-2 imagery, and optimizes logistics for emergency services.
The agent operates on the Qwen3.7-Plus model, employing a ReAct cycle to interpret user requests. To manage large volumes of geospatial data, the developers implemented a caching mechanism with a 5-minute TTL and separated data streams, sending only summaries to the LLM while routing heavy GeoJSON files directly to the frontend.
The system includes a secure Python sandbox for custom GIS analysis, protected by four layers of security: network isolation in Docker, AST filtering, restricted dunder-method access, and Linux kernel resource limits. The backend runs on Django REST Framework with PostgreSQL 16 and PostGIS 3.4.
Fire Monitor supports the Model Context Protocol (MCP), allowing integration with external development environments like Cursor or Cline. Future development plans include predictive fire spread modeling, offline capabilities using lightweight models, and voice control integration.
