CampusGuide AI — College Assistant
Django campus chatbot that answers from normalized college data first, then hybrid retrieval, calling an LLM only when needed.
Overview
CampusGuide AI for Government Degree College Ekkaghund: Excel import with data-quality reports, structured academics/admissions/campus models, hybrid FAISS/TF-IDF retrieval, and optional Groq/Gemini LLM fallback.
Problem
Students and visitors ask the same campus questions; answers live in messy spreadsheets and need a reliable assistant.
Solution
Normalize workbooks into Django models, rebuild a retrieval index, prefer deterministic lookups, then hybrid RAG + optional LLM.
Architecture
Excel import → KnowledgeSource/chunks + structured models → retriever (FAISS/TF-IDF) → services orchestration → chat UI (LLM only if needed).
Implementation
Schema-aware workbook adapters, provenance fields, golden-question evaluation command, fail-closed embeddings, Folium maps dependency, Groq/Gemini optional keys.
Results
Imported campus knowledge (hundreds of syllabus courses and chunks) with evaluation tooling and safer retrieval behavior.
Features
- Structured answers for faculty, programs, admissions, facilities, syllabus
- Workbook import with data-quality reporting
- Hybrid retrieval with lexical fallback
- Golden question evaluation
- Per-browser conversation ownership