Gaurav Chauhan

Designed and evaluated governed long-term memory for personalized LLM agents, including controlled admission, scope-aware retrieval, lifecycle management, conflict handling, and human review. Built a Confluence RAG pipeline over 178 pages and 2,362 chunks, fixed macro-ingestion gaps, and improved pre-rerank Hit@5 from 93.3% to 100%. Evaluated reranking, multi-document retrieval, generation quality, and end-to-end memory behavior. Thesis grade: 1.0.

Built an AWS SageMaker pipeline for supervised fine-tuning, evaluation, deployment, and inference of Falcon and Llama 3.2 models. Developed LangGraph agent examples and a multi-source retrieval workflow connecting GitHub, Jira, Confluence, and internal databases for contextual engineering information and GitHub-Jira traceability.

Master's thesis completed at BMW Group: Design and Evaluation of Governed Long-Term Memory for Personalized LLM Agents. Submitted and defended in August 2026 with grade 1.0. Degree completion expected in March 2027.