PolicyMind AI - RAG-Powered Company Policy Assistant

Company policies are often stored in lengthy PDF documents, making it difficult for employees to quickly find specific information. Manually searching through documents can be time-consuming and may lead to inconsistent understanding of company rules. This project explores how AI can simplify access to organizational policy information through natural language questions.

Developed PolicyMind AI, a Retrieval-Augmented Generation (RAG) application that allows users to ask questions about company policies in natural language. The system processes a company policy PDF using PyPDF, splits the extracted text into overlapping chunks, and generates vector embeddings using Sentence Transformers (all-MiniLM-L6-v2). The embeddings are stored in ChromaDB for semantic retrieval. When a user submits a question, the application retrieves the three most relevant document chunks and provides them as context to an LLM to generate a grounded answer. The application exposes REST API endpoints using FastAPI, including /health for service health checks and /ask for policy questions. Docker is used for containerization, while Langfuse and RAGAS are incorporated for observability and retrieval-augmented generation evaluation.

Delivered a functional RAG-based policy assistant prototype that converts a company policy PDF into a searchable knowledge base. The system indexes 9 document chunks and retrieves up to 3 relevant chunks per question to support context-grounded responses. The application provides API access through FastAPI and includes a separate evaluation workflow to assess answer quality and retrieval performance. Business-level productivity and cost improvements have not yet been measured in a real organizational environment.