A self-healing text classification pipeline built with LangGraph and a fine-tuned DistilBERT (LoRA) model. The system uses prediction confidence to decide whether to accept, reject, or trigger a fallback (user clarification or alternative strategy). Designed with a clean CLI, structured logging, and human-in-the-loop reliability. Demo video:

A modular, explainable recommendation pipeline leveraging multiple strategies—collaborative filtering, embeddings, and fallback logic—for robust, personalized product recommendations in real-world scenarios.

Production-ready iOS liveness detection with multi-vendor support, automatic fallback, and enterprise security. Built with Swift 6 actors and async/await.