Swarnava Dutta

Architected an end-to-end agentic RAG platform using Azure AI Foundry and Azure AI Search, generating quizzes and detecting learner skill gaps for 6k+ monthly sessions and improving detection accuracy by 37%. Built an LLM evaluation pipeline (RAGAS) that cut low-quality responses by 38% and engineered an adaptive learning engine that increased concept retention by 42% while reducing redundant stud

Optimized and scaled a supply chain copilot with LangGraph, Pinecone, and FastAPI to serve 10K+ daily queries, reducing latency by 63% and maintaining 99.9% uptime. Fine-tuned Mistral-7B with LoRA to boost contextual recall by 34%, evaluated using BLEU/ROUGE-L and retrieval metrics (recall@k), and improved multi-turn accuracy by 28% while reducing context errors by 42% using MCP-based orchestratio

Led the company’s first enterprise AI agent and production-grade multilingual RAG platform, cutting manual knowledge lookup by 43%. Instituted BLEU and NDCG benchmarks to drive measurable model performance improvements.

Orchestrated pipelines for embedding generation and lightweight text classification to enable native semantic search prototypes adopted by 13 B2B users. Improved dataset usability by 26% through pipeline and data preparation enhancements.

Implemented delta-based ingestion into Azure Data Lake for a global banking client, reducing nightly batch runtime by 95% and failures by 61%. Ensured compliance with strict financial SLAs through robust ingestion and data processing practices.

Completed a Bachelor of Technology in Computer Science and Engineering from Maulana Abul Kalam Azad University of Technology (2016–2020).