● Architected Hybrid RAG pipeline combining BM25 (Elasticsearch) and dense embedding retrieval (OpenAI Ada-002, E5-large, bge-reranker) with Reciprocal Rank Fusion, improving retrieval precision and contextual grounding by 35%

● Led a full-stack migration from Node.js to FastAPI and React to Next.js with Tailwind & shadcn UI, establishing CI/CD pipelines on GitHub and improving development speed and productivity by 40%

● Built a LangGraph-based multi-agent orchestration layer for retrieval, validation, and workflow generation, empowering users to automate app creation through natural language

Modernized the platform by refactoring 4,500+ lines of legacy code into typed, well-documented modules; implementing CI/CD to auto-lint/test/scan every PR; & parallelizing core algorithms with multiprocessing to cut execution time by 90%

● Orchestrated HIPAA-compliant data pipelines across diverse healthcare sources, ensuring 100% privacy for LLM training

● Boosted base LLM performance by 20% through RLHF on clinical question–answer pairs, leveraging GRPO and DPO techniques to enhance factual accuracy, reasoning consistency, and domain-specific reliability

● Fine-tuned specialized LoRA adapters on the aligned model and implemented dynamic adapter routing, improving question-answering accuracy and contextual adaptability by 30% across healthcare subdomains

● Engineered & optimized scalable Machine Learning models such as XGBoost, CatBoost, Random Forests to identify defects resulting in a yield improvement exceeding 0.3%

● Communicated model insights to stakeholders using Explainable AI techniques leading to company-saving of over $100,000

● Deployed web app for automated model training, resulting reduction of more than 50% in manual work hours

● Architected a suite of scalable data loaders transforming raw feeds into analytics-ready datasets, unblocking 4+ downstream teams & 20+ engineers

I worked on applying machine learning to identify defects in neuromorphic circuits. I trained and optimized classification models to detect defect patterns such as line, ring, and other structural anomalies, improving accuracy and reliability in defect detection.