BuzzAdvisor

BuzzAdvisor is a RAG-based LLM academic advisor agent designed to handle a wide range of student inquiries, including course reviews, ratings, and general information. It delivers fast, accurate, and insightful responses, ensuring students receive the information they need with ease and precision.

Type of Questions BuzzAdvisor can handle:

1. Review-Related Questions: “Could you recommend me an interesting CS course?”
   
2. Course-Specific Questions: ”What are the prerequisites and syllabus for CS 6220?”
   
3. Quantitative Questions: What are the 5 most difficult courses in CS?”
   

Environment Setup

python3 -m venv venv
source venv/bin/activate

pip install -r requirements.txt

Load Database

python3 scripts/insert_data.py
python3 scripts/insert_chroma_course_info.py
python3 scripts/insert_chroma_course_info.py

.env file

Need to attach an .env file as shown in below in the project root directory

OPEN_API_KEY=<your_api_key>
OPENAI_API_KEY=<the_same_api_key>

REVIEWS_CHROMA_PATH=./data/chroma_data_review/
COURSE_INFO_CHROMA_PATH=./data/chroma_data_course_info/
MODEL=gpt-3.5-turbo-0125

COURSE_RATING_SQLITE_PATH=data/gatech_courses.db

Running the Project

Start Backend

fastapi run backend/main.py

Start Frontend

streamlit run frontend/chatbot.py

Project Structure

• data/: stored the raw and processed data
• frontend/: the code for the Streamlit frontend
• backend/: the code for the backend LangChain architecture and FastAPI endpoints
• scripts/: web scrapping and data-processing scripts used

References

Open-Source Packages

• Chroma vector database: https://www.trychroma.com/
• SQLite relational database: https://www.sqlite.org/
• LangChain framework for agent construction: https://www.langchain.com/
• Streamlit frontend template: https://llm-examples.streamlit.app
• FastAPI sever endpoints: https://fastapi.tiangolo.com/
• HTML parsing: https://pypi.org/project/beautifulsoup4/

Dataset

Our dataset is collected through web scrapping over the

• OMS Central: https://www.omscentral.com/
• OMSCS Official Website: https://omscs.gatech.edu/current-courses

Performance Measurement

We ensembled the following 3 exsiting LLMs to evaluate our solution by comparing it with the responses generated by ChatGPT-3.5

• ChatGPT-4o: https://chatgpt.com/
• Qwen2.5: https://tongyi.aliyun.com/
• Llama3.2: https://www.llama.com/