Designed an offline evaluation protocol for comparing Query Suggestion models to reflect business OKRs

Improved null result reduction by 60% and query suggestion diversity by 30% compared to the competitive baseline while maintaining a latency bound.

Designed and implemented an OAuth2 SSO authentication and configurable RBAC-based authorization service.

Led a pilot with a customer by designing and implementing PoC, eventually onboarding as a paying customer.

Delivered a model monitoring feature enabling users to discover and debug data errors in their ML pipelines.

Optimised and automated cash and stock reconciliation processes for market securities business utilizing Hive-based data warehouse, reducing the turnaround time from two business days to less than fifteen minutes.

Developed pipelines and evaluated models for a predictive analytics tool to assist engineers on-call for major incidents using Natural Language Processing techniques.

Curated annotated datasets and set up experiments to analyze linguistic homophily in social networks resulting in publications at top conferences and workshops.