AI-POWERED FACULTY ASSISTANT

This paper presented the AI-Powered Faculty Assistant, a full-stack web application that brings five capabilities together in one platform: coordinate-based OCR timetable extraction, a hybrid rule-based and Gemini-powered chatbot, structured syllabus and topic tracking, automated contextual email notifications, and a teaching notes system. The application was built using FastAPI, SQLAlchemy, React 18, TailwindCSS, OpenCV, Tesseract, and Google Gemini 1.5 Flash. The results were encouraging across the board. The OCR pipeline achieved 86.7% accuracy on real college timetable images, outperforming the text-stream baseline by a wide margin. The notification system delivered all 47 triggered reminders over a five-day test period with zero duplicates. The chatbot classified 92% of test queries correctly, with the Gemini fallback handling every edge case cleanly. Beyond the numbers, the coordinate-based OCR approach offers a genuinely more reliable way to parse grid-format documents compared to text-stream methods. The two-layer chatbot architecture shows that deterministic and generative AI components work better together than either does alone: the rule-based layer delivers speed and predictability for everyday queries, while the LLM layer provides the flexibility needed for open-ended questions. Looking ahead, there are four clear directions for future development: building native Android and iOS applications so notifications reach faculty on their phones directly; adding multilingual support using multilingual embedding models to serve Tamil and Hindi medium institutions; integrating with LMS platforms like Moodle and Google Classroom via their REST APIs for two-way syllabus synchronisation; and implementing retrieval-augmented generation so the chatbot can answer questions grounded in the faculty's own uploaded lecture slides and textbook PDFs.

[1] A. V. Joshi and S. Kumar, “AI-Based Teaching Assistants for Educational Institutions,” IEEE Transactions on Education, vol. 66, no. 3, pp. 112–120, 2023. [2] R. Smith, “Tesseract: An Overview,” IEEE Conference on Document Analysis and Recognition, pp. 629–633, 2023. [3] M. Abadi et al., “Natural Language Processing for Domain-Specific Chatbot Systems,” Springer LNCS, vol. 13201, pp. 45–60, 2022. [4] S. R. Gundecha and R. S. Mudholkar, “Design and Implementation of a Smart Notification System for Academic Institutions,” International Journal of Computer Applications, vol. 182, no. 11, 2021. [5] P. Kochhar, S. Xiao, P. W. Lo, and J. Sun, “Automatic Classification of Software Emails,” IEEE ICSM, 2020. [6] R. Florian et al., “Named Entity Recognition through Classifier Combination,” Proc. CoNLL-2003, 2003. [7] S. Ramírez, “FastAPI Modern Python Web Framework,” https://fastapi.tiangolo.com, 2024. [8] Google DeepMind, “Gemini 1.5 Technical Report,” Google AI Blog, 2024. [9] OpenCV Documentation, “OpenCV: Open Source Computer Vision Library,” https://opencv.org, 2024. [10] N. Smith, “Linguistic Structure Prediction,” Morgan & Claypool Publishers, 2011.