Intelligent Enterprise Assistant: AI-driven Chatbot

Abstract

This paper presents the \'AI-Powered HR Assistant,\' an intelligent chatbot developed as a capstone college project aimed at transforming internal organisational support. The system incorporates a distinctive Hybrid AI architecture that merges rapid local semantic search via Sentence Transformers with the sophisticated generative capabilities of Google Gemini 2.0 Flash Exp, providing immediate and precise HR policy guidance. The system incorporates enterprise-grade security measures, featuring OTP-based authentication and a crucial \'same-user only\' policy for the generation of 16 types of official documents. It also offers advanced tools for processing large, complex PDFs (up to 50MB) for intelligent table extraction and AI-driven summarisation. This assistant is fully configurable and highly scalable, serving as a practical and secure framework for improving organisational efficiency and safeguarding sensitive data.

Introduction

The text presents an AI-Powered HR Assistant, designed to streamline organizational HR processes by providing fast, accurate, and secure responses to employee queries. Traditional knowledge management in enterprises is fragmented, leading to inefficiencies, high ticket volumes, and employee dissatisfaction. While general-purpose Large Language Models (LLMs) like ChatGPT excel at natural language understanding, they lack access to proprietary company data and are prone to “hallucinations,” making them unsuitable for enterprise HR tasks without modification.

Key Features and Innovations:

1. Retrieval-Augmented Generation (RAG):

   • Combines semantic search and generative AI to ground answers in the organization's confidential knowledge base, eliminating hallucinations.

   • Enables context-aware and authoritative responses for complex HR queries.

2. Hybrid AI Architecture:

   • Integrates deterministic rules (for security/compliance) with generative LLM dialogue.

   • Supports multi-step, agentic workflows for HR operations.

3. Advanced Document Handling:

   • Processes large PDFs (up to 50MB, 30+ pages) with table extraction, summarization, and intelligent structure analysis.

   • Automatically generates 16 types of HR documents using secure, templated workflows with strict Same-User restrictions.

4. System Architecture:

   • Frontend: Next.js 14, React 18, TypeScript, Tailwind CSS for responsive UI.

   • Backend: FastAPI with Python 3.12.0 for API handling, authentication, and middleware security.

   • AI Core: Hybrid QA Engine combining Sentence Transformers for semantic retrieval and Google Gemini 2.0 Flash Exp for generative synthesis.

   • Persistence: MongoDB Atlas for employee data, local JSON for HR Q&A knowledge base.

   • Document Stack: PyMuPDF, PDFPlumber, OpenCV for PDF processing; ReportLab for document generation.

5. Security and Compliance:

   • OTP-based MFA login and JWT session management.

   • Same-User policy enforces that employees can only generate their own documents.

   • Input validation and content filtering prevent XSS attacks and inappropriate content.

   • Full audit trail for compliance and accountability.

6. Performance Goals:

   • Low Latency: Sub-2-second response times.

   • High Availability: 99.9% uptime with redundancy and failover.

   • Scalability: Supports concurrent users with containerized, horizontally scalable architecture.

   • Error Monitoring: Tracks error rates to quickly resolve system issues.

Workflow Overview:

1. Employee logs in via OTP-based MFA.

2. HR query is sent to the AI core.

3. Semantic retrieval extracts relevant policy chunks from the local knowledge base.

4. Generative synthesis (RAG) produces accurate, grounded answers.

Conclusion

Conversational AI can be designed as a highly secure, dependable, and multipurpose enterprise platform, as the developed AI-Powered HR Assistant project effectively illustrates. The system guarantees that each response is precise and based only on authoritative organisational data by committing to a Retrieval-Augmented Generation (RAG) architecture. Additionally, the stringent Same-User Document Generation rule and the required OTP authentication create a framework of trust and compliance that cannot be compromised when managing sensitive HR data. The quantifiable advantages are evident: near-perfect availability, quick responses (less than two seconds), and less HR workload (high Self-Service Rate). This presents the Assistant as a crucial platform for revolutionising internal support rather than just an automation tool. The goal of future research will be to fully utilise the Agentic AI architecture. In order to automate intricate, multi-step organisational workflows beyond simple information retrieval and move closer to a self-sufficient digital workplace, this entails creating complex planning and execution frameworks that enable the assistant to dynamically interface with a wider range of enterprise systems (e.g., actual HRIS or ERP platforms).

References

[1] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, ?., & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30. [2] Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., ... & Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Processing Systems, 33, 9459–9474. [3] Reimers, N., & Gurevych, I. (2019). Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. arXiv preprint arXiv:1908.10084. [4] Gemini Team, Google. (2023). Gemini: A Family of Highly Capable Multimodal Models. Google DeepMind Technical Report. arXiv preprint arXiv:2312.11805. [5] Adamopoulou, E., & Moussiades, L. (2020). An Overview of Chatbot Technology. Artificial Intelligence Applications and Innovations, 584, 373–383. Springer. [6] Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., ... & Fung, P. [7] (2023). Survey of Hallucination in Natural Language Generation. ACM Computing Surveys, 55(12), 1–38. [8] Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. Proceedings of NAACL-HLT, 4171–4186. [9] Weizenbaum, J. (1966). ELIZA—a computer program for the study of natural language communication between man and machine. Communications of the ACM, 9(1), 36– 45. [10] Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M. A., Lacroix, T., ... & Lample, G. (2023). LLaMA: Open and Efficient Foundation Language Models. arXiv preprint arXiv:2302.13971. [11] Gao, Y., Xiong, Y., Gao, X., Jia, K., Pan, J., Bi, Y., ... & Wang, [12] H. (2023). Retrieval-Augmented Generation for Large Language Models: A Survey. arXiv preprint arXiv:2312.10997. [13] Ramirez, S. (2023). FastAPI: Building Data Science, Web, and RESTful Applications in Python. O\'Reilly Media. [14] Vercel. (2024). Next.js 14 Documentation: App Router and Server Actions. Available at: https://nextjs.org/docs. [15] Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., ... & Zhang, Y. (2023). Sparks of Artificial General Intelligence: Early experiments with GPT- [16] arXiv preprint arXiv:2303.12712. [17] Wei, J., Wang, X., Schuurmans, D., Bosma, M., Chi, E., Le, Q., & Zhou, D. (2022). Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. Advances in Neural Information Processing Systems, 35, 24824–24837. [18] Karn, S. K., & Ulanova, L. (2023). Enterprise Search with Large Language Models: Opportunities and Challenges. Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. [19] ReportLab Inc. (2024). ReportLab PDF Generation Library Documentation. Available at: https://www.reportlab.com/docs/. [20] Wallace, E., Feng, S., Kandpal, N., Gardner, M., & Singh, [21] S. (2019). Universal Adversarial Triggers for Attacking and Analyzing NLP. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. [22] Richardson, L. (2023). Microservices Patterns: With examples in Python and FastAPI. Manning Publications.

Copyright

Copyright © 2025 Prof. S. V. Chaudhari, Devyani Suresh Deore, Ashwini Anil Nikumbh , Khushbu Arun Jain, Mansi Anil Badgujar. This 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.