What is it?
An AI-powered virtual student assistant that answers questions directly from the university knowledge base using Retrieval-Augmented Generation.
The problem
The university had a large knowledge base that students rarely explored directly. Instead, students went to campus ambassadors to ask repetitive questions.
The idea
Build a RAG-based chatbot that could answer student questions directly from the university knowledge base, making information accessible 24/7.
My role
Designed and built the entire system: conversational interface, vector database, RAG pipeline, and OpenAI API integration.
How it works
Students ask questions through a SvelteKit chat interface. The system retrieves relevant context from a Weaviate vector database and generates grounded answers using OpenAI APIs.
Key decisions
- — Chose RAG over fine-tuning for accuracy with university-specific data
- — Weaviate for vector storage and semantic search
- — SvelteKit + Flask architecture for modern frontend with Python AI backend
- — Grounded responses to prevent hallucination on university facts
What I learned
- — First experience identifying a practical problem and applying emerging AI technology
- — RAG is powerful for domain-specific knowledge bases
- — University interest validated the problem, even if the project was not continued
- — Foundation for all subsequent AI work