v1.0.0

First tagged release of the cloud-based, retrieval-augmented drone framework for early forest fire detection and response planning.

A drone-side YOLO detector flags fire or smoke and reports an incident to a cloud API. The API stores the snapshot and the incident, sends an immediate alert to responders, and only then runs a RAG pipeline that grounds an LLM response plan in wildland-fire SOP documents, with everything streaming to a live operator dashboard.

Measured results (from the README and results/)

Measurement	Result
Fire detection, FireNet validation (90 images)	mAP@0.5 0.733, precision 0.766, recall 0.739. The 0.88 target was not met
Detect, alert and plan pipeline (mock LLM)	about 73 ms mean
Telemetry throughput, 10 simulated drones	about 198 messages/s, 0 errors

Metrics that have not been measured are marked PENDING in the results folder rather than estimated.

Stack

YOLOv8n edge detector, FastAPI gateway with REST and WebSocket ingestion (plus an optional MAVLink bridge), sentence-transformers embeddings with a FAISS index, Gemini Flash for plan generation (a deterministic mock runs offline and in CI), React and Leaflet dashboard, and Terraform for Azure Container Apps.

Changes in this release

• The Azure deploy workflow is now manual-only: the container registry it pushes to has been torn down, so running it on every push only produced a failing check. The push trigger is kept as a comment for when the registry returns