On a shopping assistant for a 12,730-product catalogue, I built a LangGraph agent and FastAPI service with hybrid retrieval and customer preference memory. Reranking raised right-product-first results from 23% to 43%.
I designed a code-graded evaluation for the agent, where a prompt update passed 12 of 12 tasks, compared with 7 previously. I also added guardrails, tracing, monitoring, automated tests, and CI.
My work also includes fine-tuning Stable Diffusion with LoRA for text-to-floor-plan generation, published in Machine Graphics & Vision, and a quantum machine-learning pipeline for EEG classification, published at the Neurogenomics 2024 Conference. I currently teach Computer Science at Culture and Science City.

