Work History

Built an enterprise inference routing framework in front of LiteLLM so user requests could be routed across multiple model providers based on task capability, predicted failure risk, and business constraints.

Designed a hybrid routing approach that combined a lightweight capability classifier, model-specific error maps, and deterministic business rules, improving routing accuracy by 20% while reducing average inference cost by 15%.

Owned the open-source Innovation Hub repository with 15 AI blueprints, independently built 4 of them, and handled updates, fixes, and feature work across the broader platform.

Built and maintained AI workflows across OpenAI, VLLM, Ollama, and Intel OPEA so the same blueprints could be benchmarked across different model and serving environments.

Built backend systems for an inventory management platform that handled product inventory, checkout, sales tracking, inventory history, dashboard reporting, and reorder recommendations for small and regional stores.

Optimized product search across 100,000+ records using composite indexes, filtered queries, and pagination, reducing average search latency from around 1.2 seconds to under 120 milliseconds.

Built the backend data layer for a forest-fire prediction research platform, transforming 100K+ daily geospatial and environmental records into structured PostgreSQL datasets for model training, prediction review, and dashboard analytics.

Redesigned PostgreSQL tables around location, timestamp, weather attributes, fire-risk scores, and model outputs, adding query-focused indexes to support faster filtering, aggregation, and comparison across regions and time periods.