RAG-Based MCA AI Agent for Document-Grounded College Information Assistance Using Retrieval-Augmented Generation, Vector Search and a Locally Hosted Large Language Model

This paper presents the design and implementation of a Retrieval-Augmented Generation (RAG)-based MCA AI Agent for document-grounded question answering over college and student policy documents. The system uses PyMuPDF for PDF text extraction, approximately 1000-character chunks, Nomic-Embed-Text for generating 768-dimensional embeddings, and Pinecone for semantic vector search. At query time, the system retrieves the three most relevant document chunks and provides them as context to Llama 3.2 running locally through Ollama. A React.js frontend and FastAPI backend provide the conversational interface. The paper describes the system architecture, methodology, implementation, document-refresh workflow, and practical deployment considerations, including stale vectors and local inference latency. It also outlines an evaluation framework for comparing a standalone LLM with the proposed RAG pipeline using retrieval relevance, answer correctness, faithfulness, and latency. The current work is presented as a system implementation and preliminary study. A controlled quantitative evaluation has not yet been completed, and no numerical accuracy improvement is claimed. Keywords: Retrieval-Augmented Generation, Large Language Models, Semantic Search, Vector Database, Pinecone, Llama 3.2, Ollama, FastAPI, React.js, Document-Grounded Question Answering.