Prototype a real-time behavioral risk detection pipeline over existing CCTV camera feeds.
Study a multi-agent LangGraph architecture that splits perception, reasoning, and intervention into separate agents.
Train and evaluate a BiLSTM classifier for temporal behavior sequences from pose-tracking data.
Build a live alert dashboard with WebSocket-based escalation for public safety monitoring.
| abishek-kk/railmind-ai | williamsmithedward/xlide_vscode | yucheng1554439/archlens | |
|---|---|---|---|
| Stars | 41 | 41 | 41 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | hard | moderate | moderate |
| Complexity | 5/5 | 2/5 | 3/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires an edge device (Jetson-class hardware suggested) for the vision pipeline plus a cloud backend with PostgreSQL and Redis, involves training or loading multiple ML models.
RailMind AI is a safety monitoring system built for railway stations. It works by connecting to existing CCTV cameras and analyzing the video feed in real time, without requiring any new hardware to be installed. The system watches for specific behaviors that may indicate a passenger is in danger or that a crime is about to occur, and sends an alert to nearby staff before the situation escalates. The two main problems the system targets are suicide prevention and theft. For suicide risk, it tracks whether a passenger spends extended time near the edge of the platform, paces repeatedly, or shows signs of withdrawal from other people. For theft, it looks for individuals who follow others too closely, make repeated contact with strangers, or move in coordinated patterns near crowded areas. Risk is scored on a scale from 0 to 100, combining how long a behavior has persisted, how close to the platform edge the person is, and other contextual factors. The technical pipeline runs in two stages. On an edge device installed at the station, a computer vision model scans each video frame to identify and track people, then estimates their body posture. Those observations feed into a type of neural network called a BiLSTM, which examines sequences of behavior over 30-second windows to classify what each tracked person is doing. That classification is then sent to a cloud server where three AI agents, called Perception, Reasoning, and Intervention, evaluate the risk and decide whether to trigger an alert. Alerts reach railway staff through a live dashboard with sub-500-millisecond latency. If a high-risk alert is not acknowledged within 60 seconds, the system escalates automatically. The platform does not use facial recognition and does not store biometric data, which keeps it compatible with privacy regulations. The project was built for a hackathon called FAR AWAY 2026. It is written in Python and TypeScript, and uses FastAPI for the backend, React for the dashboard, and LangGraph for the AI agent logic.
An agentic AI system that turns existing railway CCTV feeds into real-time alerts for suicide risk and pickpocketing, without new hardware or facial recognition.
Mainly TypeScript. The stack also includes Python, TypeScript, FastAPI.
Use freely for any purpose, including commercial use, as long as you keep the copyright and license notice.
Setup difficulty is rated hard, with roughly 1day+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.