safety-ai

Multi-modal content moderation system. Detects toxic text and unsafe images in real-time using BERT and Computer Vision. Designed for high-performance safety governance.

Python package to accelerate research on generalized out-of-distribution (OOD) detection.

Open-source safety and cognitive control infrastructure for physical AI agents — robots, LLM agents, autonomous systems.

Code for the paper "A Data-Driven Measure of Relative Uncertainty for Misclassification Detection"

A custom YOLOv8 helmet-detection model trained for 100 epochs using a Roboflow dataset. Includes scripts for real-time image and video detection with bounding boxes, confidence scores, and object tracking with persistent IDs for accurate rider safety monitoring.

Scalable, asynchronous multimodal content moderation pipeline with CV, LLM reasoning, and human-in-the-loop review.

AI response safety/ethics/accuracy checker - by A0 (currently not maintained)

🏥 MedAssist — Multi-Agent Clinical & Research Intelligence System. Advanced RAG + Safety Layer for medical Q&A with cited, hallucination-free answers. Built with LangChain, ChromaDB, Groq, and Streamlit.

AI-powered ambient distress detection prototype for proactive personal safety with future wearable integration.

Real-time person-tracking with YOLOv8, zone-dwell analytics and automatic alerts.

Real-time fall detection using YOLOv8 pose estimation, geometric posture analysis, timestamped alerts, and PyTorch versus ONNX benchmarking.

A robust, real-time Face Mask Detection system built with Python, OpenCV, and TensorFlow. Features automated environment setup, FPS monitoring, and fail-safe execution.

Safe multi-modal autonomous driving with cross-attention perception and formally verified QP runtime enforcement shield.

Agentic AI system for Nordic wind-turbine manufacturing using AWS Bedrock and Claude, reducing downtime, enforcing safety, and preventing high-cost mechanical failures.

AI-powered autonomous sentinel and safety perception system

Real-time computer vision safety monitoring system using OpenCV and PyTorch for industrial hazard detection

AI-powered safety equipment monitoring system to ensure workplace compliance through real-time detection.

Reinforcement learning framework for training autonomous F1 racing agents with CBF safety guarantees, real F1 track data, MuJoCo physics simulation, and multi-agent racing support

Real-time adaptive learning system for autonomous vehicles using NVIDIA Cosmos Reason2-2B. Generates safety rules from edge cases, achieving 1500x speedup after learning. Built for NVIDIA Cosmos Developer Competition 2026.