Developed production-grade RAG pipelines using LangChain, PostgreSQL, and PGVector to optimize clinical data retrieval for healthcare providers, reducing query response time by 60%. Built multi-agent LLM validation frameworks with AutoGen and stateful memory for clinical document review, processing 50K PDFs and reducing compliance audit preparation time by 90%. Implemented comprehensive RAG monitoring dashboards tracking hallucination detection and query distribution drift across 50K clinical interactions, improving model reliability by 15%. Created scalable preference learning pipelines with human-in-the-loop validation for DPO and RLHF, improving factual consistency by 10% and accelerating dataset preparation from 14 days to 1 day. Rewrote model training and inference pipeline codebase with custom PyTorch fused ops and vLLM optimizations on Nvidia A100 and RTX 6000 Ada clusters, reducing training and evaluation cycles from 4 hours to 1 hour.

Software Engineering Intern, Quality Analytics at Boehringer Ingelheim

December 1, 2024 - October 8, 2025

Deployed production ML models via APIs using LlamaIndex, OpenAI GPT-4, and FAISS for pharmaceutical quality control, reducing manual survey analysis from 1 week to 5 minutes. Built Master Data Management platform leveraging AWS Bedrock, SageMaker, and Step Functions with vector search capabilities, saving quality analysts 30 hours weekly on audit preparation. Performed big data processing in TrackWise Oracle DB by designing ETL pipelines with PySpark, Databricks, SQL, improving data processing speed by 100%, and delivering 6 Power BI dashboards for daily tracking. Conducted rigorous model evaluation and A/B testing for multilingual embedding models in QA systems, improving global document retrieval accuracy by 20% across international manufacturing sites.

Supply Chain Analyst at Ecolab

July 1, 2023 - October 8, 2025

Designed and shipped a supply chain forecasting and replenishment platform, built Spark ETL pipelines writing to Hive Parquet tables, trained models, served FastAPI endpoints, reducing stockouts 19%, saving planners 25 hours weekly. Cut inventory write-offs by $2.3M annually across Target, BlueYonder, and 3PL clients by building demand forecasting solutions combining ARIMA, exponential smoothing, and LSTM models for 500 SKUs, lifting accuracy by 20%. Optimized batch processing pipelines and model deployment infrastructure, saving supply chain planners 25 hours weekly through automated forecasting and replenishment recommendations. Analyzed a regional demand spike in Snowflake using SQL by joining sales, inventory, weather data, identifying weather related port closure and created Cruise Scorecard Dashboard to alert Logistics team, cutting response time by 50%.