AI-Based University FAQ Chatbot
A full-stack university FAQ chatbot that compares a TF-IDF and cosine-similarity baseline with an LLM-RAG pipeline using LangChain, FAISS, Hugging Face embeddings, and Groq.
Conversation continuity
An interface concept for revisiting previous university questions and continuing earlier conversations.
Overview
A full-stack university FAQ chatbot that compares a TF-IDF and cosine-similarity baseline with an LLM-RAG pipeline using LangChain, FAISS, Hugging Face embeddings, and Groq.
The Problem
Traditional university websites often force students to hunt through static pages and dense PDFs for basic answers. While keyword search helps, it struggles with conversational questions or nuanced context like scholarship eligibility.
The Approach
I developed a dual-pipeline architecture to solve this. The first is a traditional TF-IDF baseline that matches exact keywords. The second is an advanced RAG (Retrieval-Augmented Generation) pipeline using FAISS and Hugging Face embeddings, passed to the Groq API (Llama 3.1 8B Instant) to generate natural, conversational answers.
AI response workflow
The project evaluates how a traditional keyword-based retrieval pipeline and a modern retrieval-augmented generation approach can support clearer responses to university-related questions.
My Contribution
I built the entire end-to-end system: parsing the university data into vector embeddings, engineering the FAISS retrieval logic, wrapping it in a FastAPI backend, and building a responsive React/Vite web interface.
Key Capabilities
- React/Vite responsive chat interface
- FastAPI backend architecture
- Traditional TF-IDF keyword baseline
- LLM-RAG pipeline using LangChain
- FAISS vector store for semantic retrieval
- Hugging Face embeddings integration
- Groq API (Llama 3.1 8B) for natural response generation
Technology Stack
Project Context
Academic project exploring NLP and GenAI architectures.