Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{197132,
        author = {Guttula Satish and Garaga Sri Pushpa Latha and Nandyala Himasri Chakravani and Bokka Hema Siri Chandana and Budharouthula Keerthi Sri and Dr. Yalla Venkat},
        title = {Enterprise Policy Assistant: A Retrieval-Augmented Generation System for Semantic Enterprise Policy Retrieval},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {5777-5782},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=197132},
        abstract = {Enterprise organisations manage extensive internal policy documentation spanning Human Resources, Information Technology, Legal, and Compliance domains. Retrieving accurate information from these documents remains a time-consuming and error-prone task when relying on traditional keyword-based search systems. Furthermore, standalone AI tools often lack enterprise security, auditability, and architectural rigour. This project presents an Enterprise Policy Assistant built using Retrieval-Augmented Generation (RAG) and vector similarity search to deliver accurate, context-grounded responses from enterprise policy documents. The system employs an N-Tier enterprise architecture comprising a React.js with TypeScript frontend, a FastAPI-based REST backend, a PostgreSQL relational database, and a Qdrant vector database for semantic retrieval. OpenAI Large Language Models are invoked exclusively after relevant policy sections are retrieved, significantly reducing hallucination risks. Security is enforced using JWT-based authentication and Role-Based Access Control (RBAC). The infrastructure runs using rootless Podman containerisation, ensuring an enhanced security posture with zero privilege escalation risks. The RAG pipeline ingests policy documents (PDF, DOCX, TXT), splits them into overlapping text chunks (1300 characters, 150 overlap), generates 1536-dimensional vector embeddings using OpenAI text-embedding-3-small, and stores them in a Qdrant collection. On each user query, the system performs cosine similarity search (Top-K 10, threshold 0.4), constructs a strict retrieval-grounded prompt, and generates a source-cited response using the gpt-5.4-nano model. All query–response pairs, source documents, retrieval scores, and token usage persisted in PostgreSQL audit logs. Evaluation demonstrates API latency under 200ms (excluding LLM generation), retrieval relevance exceeding 90%, and hallucination rates below 5%. The system demonstrates how modern AI techniques can be responsibly integrated into enterprise-grade software while maintaining architectural rigour, compliance, and scalability.},
        keywords = {FastAPI, OpenAI, Podman, Qdrant, PostgreSQL, REST, Compliance domains, Enterprise Policy, Evaluation.},
        month = {April},
        }