At BIRMARKET, I developed the retrieval layer for a live RAG customer support chatbot, grounding generated answers in internal business documentation. I also designed and deployed a personalized ranking system, validating its impact through online A/B testing.
I improved BIRMARKET’s similar-items recommendations with Siamese Network embeddings and FAISS; the approach outperformed the previous solution in evaluation and was adopted in production. I also designed, implemented, deployed, and maintained ML workflows across Azure services.
On my Booking AI System project, I built a conversational accommodation search system that retains preferences across turns and retrieves and ranks listings. I created repeatable evaluation pipelines, achieving ~86% resolution accuracy and ~98% top-1 ranking accuracy, and added observability for latency, usage, cost, fallbacks, and API failures.
