PSDSARCSAR Monitor
Agentic UAVs · Search & Rescue · Public Safety

Drones that think while they watch.

PSDSARC is an LLM-driven monitoring system for aerial drone feeds. YOLO detects, a vision LLM reasons, and an agent decides — calling alerts, escalating to teams, and logging incidents in real time. Built for Hajj-scale crowds and beyond.

Live Monitor

Annotated MJPEG feed with YOLO bounding boxes, severity badge, and the agent's reasoning surfaced frame by frame.

Agentic Reasoning

Vision LLM (GPT-4o / Claude / Gemini) classifies each frame, calls tools (send_alert, escalate_to_team, log_incident) and respects operator overrides.

Mission Presets

Hot-swap between five missions — Pilgrim SAR, Pilgrim Security, Combined, Find Missing Person, General Surveillance — with no restart.

Alert Archive

Every alert is persisted with the captured frame and delivery status for email (SendGrid) and Telegram. Ask follow-up questions in natural language.

📡

Source Agility

Local webcam, video file, RTSP/RTMP, or DJI via MediaMTX — all hot-swappable from the dashboard. Upload a clip and start playing in one click.

🛡

Auditable

Every tool call, every incident, every operator directive is logged to SQLite. Replay, audit, or fine-tune from real data.

Built on

YOLO11 GPT-4o · Claude · Gemini Flask + SSE OpenCV + FFmpeg MediaMTX SQLite SendGrid · Telegram

A PSDSARC research demo

Developed at Alfaisal University in collaboration with ScaleX Innovation as part of the PSDSARC programme on agentic AI for autonomous UAVs in defense and emergency response.