GeezerVison

- Camera CRUD with go2rtc RTSP stream registration
- Incident model with type/severity classification (traffic_accident, fight, etc.)
- VLM analysis endpoint: snapshot capture → Celery job → VLM → incident record
- LLM agent query endpoint with DB context injection
- Shift/daily report generator via LLM agent
- WebSocket real-time alert stream via Redis pub/sub
- MinIO snapshot storage service
- PostgreSQL + asyncpg async ORM (SQLAlchemy 2.0)
- Celery worker for async VLM jobs
- Docker Compose stack: backend, worker, db, redis, minio, go2rtc

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
Claude-Session: https://claude.ai/code/session_01DXvYppU6bAHzS9R8HbTHdN

- Dark-theme dashboard with stat cards, severity pie chart, incident type bar chart
- Multi-camera live grid (2/3 col toggle) with per-cell VLM analysis trigger
- Incident table with filter, resolve/false-alarm actions
- AI Agent chat page with quick-suggestion prompts and DB context display
- Shift report generator with time window + district filter and TXT download
- Camera manager (CRUD, toggle active, delete) with go2rtc registration
- Real-time alert ticker bar via WebSocket with auto-reconnect
- Full TypeScript, Tailwind dark theme, Recharts for charts
- Dockerfile (multi-stage) + docker-compose frontend service

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
Claude-Session: https://claude.ai/code/session_01DXvYppU6bAHzS9R8HbTHdN

7 analysis modes with tailored VLM prompt templates:
- traffic_accident, crowd_anomaly, suspicious_behavior
- wrong_way, road_obstruction, fire_smoke, full_scan

Features:
- Sensitivity levels (low/medium/high) adjusting VLM strictness
- Language support: English, Burmese, or bilingual output
- Camera location context injection into prompts
- YOLO pre-filter: only sends frames to VLM when trigger objects detected
- Frame throttle (configurable FPS) to manage VLM token cost
- Full Aegis JSONL stdin/stdout protocol compliance
- Zero extra Python dependencies (stdlib urllib only)
- Standalone test script for prompt validation per mode
- deploy.sh with prompt module sanity check

Co-Authored-By: Claude Sonnet 4.6 <[email protected]>
Claude-Session: https://claude.ai/code/session_01DXvYppU6bAHzS9R8HbTHdN