SSIT Campus Regulations Assistant
Meowfessional Coders · Hacktoberfest Hack Day · Theme 2A: RAG Application with Gemma

Students at Sri Siddhartha Institute of Technology (SSIT), Tumakuru often struggle to find clear answers about college rules. The information is spread across a long handbook and many branch-wise syllabus documents, and general AI chatbots can give confident but wrong answers about a specific college. The Campus Regulations Assistant solves this with a Retrieval-Augmented Generation (RAG) chatbot that answers students’ questions using only SSIT’s official documents, and cites the document and page for every answer.

The assistant draws on the SSIT Hand Book cum Work Planner 2025-26 and the syllabus documents of every branch and year: CSE, CSE (Cyber Security), CSE (Data Science), ECE, Mechanical, Civil and the common first year, plus the M.Tech. programmes. These documents were cleaned and divided into over 3,600 searchable passages. The handbook is split rule by rule, so each regulation stays complete, and each syllabus is split course by course and unit by unit. Every passage is tagged with its document, page, programme, branch, year and semester. Information that existed only as images, such as university notifications and scheme tables, was extracted with OCR.

Students can ask questions in everyday language: minimum attendance, passing marks, grade points, carry-over rules, AICTE activity points, internships, scholarships, anti-ragging rules, holidays, the subjects in a semester, or who the HOD of a department is. The assistant replies with a short, structured answer: the direct answer, the key details, related rules where programmes differ, and the exact source. Students can also ask for a document itself, such as “send me the ECE 2nd year syllabus”, and receive the official PDF. Questions the documents don’t cover, such as hostel timings or re-evaluation deadlines, are declined rather than guessed, and the student is directed to the right office.

The system combines meaning-based search (the all-MiniLM-L6-v2 embedding model) with keyword matching to find the most relevant passages. Students can filter results by their own programme, and every answer comes with the passages it used, so they can verify it. Google’s open-weight Gemma 4 model generates grounded answers from that retrieved evidence. The interface is built with Gradio, and the whole project uses only free and open-source tools.

Tech stack: Python · pypdf · RapidOCR · Sentence-Transformers (all-MiniLM-L6-v2) · Gemma 4 (Google AI Studio) · Gradio