Hire Dharan Y. - RAG Engineer

Contributing to the digital transformation of USP-NF by developing LLM-driven Agentic RAG systems over large-scale legacy pharmaceutical and regulatory knowledge bases, enabling structured access to complex domain standards and regulatory knowledge used across global pharmaceutical workflows. ~ Designing and optimizing modular Agentic RAG pipelines using LangChain and LangGraph, enabling stateful and asynchronous agent workflows, tool orchestration, query routing, hierarchical retrieval, metadata-aware search, and advanced chunking strategies tailored for domain-grounded reasoning. ~ Experimenting with and evaluating multiple LLMs, vector embedding models, and retrieval configurations, including chunking granularity, semantic indexing approaches, memory/state handling strategies, and retrieval evaluation techniques to improve grounding accuracy, response reliability, and observability across enterprise knowledge workflows. ~ Developing agent workflows leveraging ReAct-style reasoning, async tool calling, and contextual state management, while evaluating multiple orchestration frameworks and workflow patterns for scalable agent automation. ~ Applying LLM-based automation beyond agentic systems by supporting AI-driven data engineering and analytical workflows, integrating LLM reasoning into structured data pipelines and enterprise processing tasks. ~ Engineering containerized GenAI services using FastAPI and Docker, enabling scalable deployment on AWS EKS in collaboration with enterprise cloud operations teams for production-grade AI delivery. ~ Working with complex proprietary regulatory and legacy datasets, transforming structured and unstructured content into context-aware enterprise knowledge systems.

Delivered AI-driven feature-extraction workflows leveraging state of art computer vision models (U-Net++, DeepLabV3+, SegFormer) for aerial/satellite imagery, 3D LiDAR point clouds, and vehicle perception data (VPD) ~Designed and orchestrated an automated AI annotation pipeline using Anthropic’s Claude on AWS Bedrock, reducing manual labeling effort by 60% and delivering 41 geospatial features to a global navigation and mapping provider within 90 days. ~Architected a multilingual, HIPAA-compliant clinical sentiment analysis system for audio/video data on AWS (EC2, S3, MLflow). Integrated Whisper for transcription, ClinicalBERT for sentiment analysis, and DeepFace for facial emotion detection—achieving accuracy comparable to AWS’s offerings at half the API cost for a Fortune 500 pharmaceutical client. ~Enhanced stock-prediction accuracy by 15% for a leading automotive tech provider through the deployment of a Prophet-powered forecasting and recommendation system for parts dealerships. ~Contributed to the development and maintenance of a library of 15+ ML models (traditional ML, computer vision, NLP) applied across healthcare, automotive, fintech, and mapping use cases. ~Automated MLOps pipelines to ensure scalable, production-grade ML deployment. ~Collaborated closely with cross-functional teams (GIS engineers, project managers, annotators) to align technical solutions with business needs and provided mentorship to junior engineers. ~Designed and implemented a Retrieval-Augmented Generation (RAG) pipeline for a Microsoft-first client, integrating Azure OpenAI Embeddings, Cognitive Search, and Cosmos DB (VectorDB) to enable Copilot-driven knowledge retrieval across SharePoint, Teams, and Outlook; enforced role-based compliance using Microsoft Purview security trimming.