Developed a state-machine conversation framework, enabling rich UI interactions and integrating a structured expert-in-the-loop feedback system for continuous improvement of retrieval coverage and answer quality.

Built a metadata-aware Pre-Ranker using feature fusion to optimize candidate selection, improving Recall@30 while reducing inference cost.

Designed and deployed an Enterprise File Search engine, combining fuzzy, dense, and sparse retrieval with metadata filters, adding capabilities like file summarization and conversational querying.

Built an unsupervised topic-insights system that surfaced product-relevant themes from customer reviews/conversations using clustering experiments.

Constructed taxonomy trees using co-occurrence graphs and ontologies, then pruned and mapped model-generated topics.

Built a digital-asset valuation model integrating rarity metrics and liquidity curves, achieving better accuracy than baseline heuristics.