• Developed a Related Query Suggestions component for multiple Adobe Products using Probabilistic Models and Knowledge Graphs

• 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.

• Optimised ML Inference servers to use Parquet files effectively and reduced memory usage by up to 85%

• 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.

• Proposed and drove the implementation of an internal MLOps platform and set up inference pipelines for problems related to Trade Settlement Analysis and Counterparty Risk Anomaly Detection.

• 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.

• Collaborated with researchers from Bloomberg Research, NY and ASU on research problems in Applied Multimodal Learning.

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