BRACU CSE Thesis Chatbot
A retrieval-augmented academic assistant that helps BRAC University CSE students find thesis and course-prerequisite information through natural-language conversations.
Overview
BRACU CSE Thesis Chatbot is a focused question-answering assistant for BRAC University CSE students. It combines a conversational Streamlit interface with semantic retrieval from a persistent academic knowledge base, helping students find thesis and course-prerequisite information through natural-language questions.
Features
- Answers questions about CSE thesis eligibility, registration, supervision, and related procedures
- Retrieves the three most relevant records for each question
- Maintains the active chat history for follow-up questions
- Responds in English or Bangla according to the user’s language
- Uses a persistent 56-record academic knowledge base
- Presents the workflow through a focused browser-based chat interface
- Detects an explicit conversation-ending signal and disables further input for that session
Problem
Information about thesis eligibility, registration, supervision, deadlines, and course prerequisites can be spread across documents and web pages. Students may need to search several sources or ask repeated follow-up questions before finding guidance that fits their situation.
Solution
The application brings project-specific academic records into one searchable conversational interface. For each question, it retrieves the three most relevant records from a persistent ChromaDB collection, combines those passages with system instructions and the current conversation history, and uses a language model to generate a grounded response.
Architecture
Interface: A Streamlit chat application stores the active conversation in session state and sends each question to the retrieval layer.
Retrieval: A persistent ChromaDB collection returns the three most relevant thesis, internship, and course-prerequisite records.
Generation: The original system used Llama 3 70B through the Groq chat-completions API. It was later migrated to openai/gpt-oss-120b after Groq removed support for the original model. Retrieved passages, system instructions, and conversation history are assembled into each prompt.
Data preparation: Python, CSV processing, and a Jupyter Notebook format 56 source records for semantic search.
Results
The completed prototype supports multi-turn thesis guidance in English and Bangla over 56 indexed academic records. The supplied walkthrough demonstrates initial prerequisite questions, a more specific CGPA follow-up, and a Bangla restatement within one conversation.
Challenges
The assistant’s intentionally narrow knowledge base can become outdated when university policies or deadlines change, so policy-sensitive answers still need verification against official BRAC University sources. Retrieval quality depends on the indexed records, and generation uses the external Groq API, which processes the question, relevant passages, and included conversation history.