Dec 2025 – Present · 5 mos · Bengaluru · Remote

• Designed and delivered an end-to-end Retrieval-Augmented Generation (RAG) architecture for healthcare document intelligence, enabling accurate, explainable, and context-aware question answering over large-scale unstructured medical and operational datasets.
• Built scalable data ingestion and preprocessing pipelines using tools such as PyPDFLoader, custom document parsers, and text normalization frameworks to handle diverse healthcare content.
• Implemented advanced document chunking strategies (recursive, semantic, and metadata-aware chunking) to preserve clinical and regulatory context, improving downstream retrieval accuracy and answer quality.
• Generated semantic embeddings using state-of-the-art embedding models (OpenAI / Azure OpenAI embeddings) and integrated them with vector databases such as ChromaDB and Azure Cognitive Search for efficient similarity and hybrid retrieval.
• Designed hybrid retrieval pipelines combining dense vector search, keyword-based filtering, and metadata constraints to significantly reduce hallucinations and improve factual grounding in healthcare responses.
• Developed an Agentic AI framework with a multi-agent architecture, where autonomous agents performed specialized roles including document retrieval, domain reasoning
• Integrated large language models (GPT-4, GPT-4o, GPT-4-mini) using controlled prompting strategies, system instructions, and safety guardrails to ensure accuracy, traceability, and alignment with healthcare domain requirements.
• Implemented Responsible AI (RAI) and governance mechanisms, including source citation enforcement, confidence scoring, rule-based validation, fallback handling, and response auditing to meet healthcare compliance and risk standards.
• Designed comprehensive evaluation frameworks using retrieval metrics (precision, recall), LLM-based evaluation techniques (G-Eval), and human-in-the-loop review processes to continuously monitor and improve system performance.