Building AI systems that hold up in the real world.

Designed and deployed multi-stage RAG with chunking strategies, hybrid retrieval, and reranking to reduce hallucination in production.

Built an automated eval pipeline covering test-case generation, multi-dimensional scoring, regression detection, and human-in-the-loop feedback.

LLM-powered system that extracts key client information from meetings and generates structured summaries, cutting manual post-meeting workload.

Led full model lifecycle — training, deployment, monitoring — for churn prediction; converted batch model to real-time inference service.

Used causal inference and A/B testing to measure the impact of product and algorithm changes on advertising revenue and user engagement.