During February 2026, Prajwal Moras developed a fully local, offline retrieval-augmented generation notebook for the run-llama/llama_index repository. Leveraging Python, machine learning, and data science skills, Prajwal implemented an end-to-end RAG pipeline using Llama3.2:3b for language modeling, nomic-embed-text for embeddings, and ChromaDB for vector storage with caching. The notebook included a config.yaml and sample dataset to ensure deterministic, reproducible runs, and achieved a 100% retrieval hit rate with an MRR of 0.85. Prajwal also demonstrated failure modes and integrated guardrails to reduce hallucinations, focusing on maintainability, offline reproducibility, and user safety in restricted environments.