LLM-Driven Study Abroad Consulting System

This project is a document-augmented question answering system combining a React frontend with a Flask backend. It supports real-time conversational interaction with a local large language model (LLM), document upload and semantic retrieval, and integrates a study abroad encyclopedia along with a curated QA dataset to enhance answering capabilities.

---

✨ Core Features

• Real-time Chat Interface
  ChatGPT-style multi-turn conversation with instant feedback.

• Document Upload and Parsing
  Upload PDF and Word documents; the system automatically parses and integrates their content into the retrieval process.

• Dense Semantic Retrieval (RAG)
  Supports retrieval from three sources: user-uploaded documents, a built-in study abroad encyclopedia, and a curated study abroad QA dataset.

• Local Large Language Model Inference
  Efficient local inference with 4-bit quantization (BitsAndBytes), supporting large models such as Qwen2.5-8B-Chat.

• Multi-Source Knowledge Fusion
  Dynamically retrieves and fuses multiple knowledge sources to build a context-aware prompt, delivering highly relevant answers.

---

🗂️ Project Structure

study-abroad-rag-system/
├── README.md
├── .gitignore
├── backend/
│   ├── app.py
│   ├── requirements.txt
│   ├── uploads/
│   ├── model/
│   ├── data/
│   │   ├── uk_study_guide.txt
│   │   └── reconverted_prompt_completion.jsonl
└── frontend/
    ├── package.json
    ├── tsconfig.json
    ├── src/
    │   ├── App.tsx
    │   ├── components/
    │   ├── types/
    └── public/

---

🚀 Quick Start Guide

This guide provides step-by-step instructions to set up and run the Study Abroad RAG System locally.

The Python version we used for development is 3.10.16

1. Clone the Repository

git clone https://github.com/GoDToD/ISAPA.git
cd ISAPA

2. Set Up the Backend (Flask)

cd backend
python -m venv env

Download and unzip fine-tuned model through the link

https://drive.google.com/file/d/1WS0EvoqH2-_glmy8LkwcTHG_iayEa3kg/view?usp=drive_link

then put it in backend/model folder.

Activate the virtual environment:

• On Windows:

  env\Scripts\activate

• On Linux/MacOS:

  source env/bin/activate

Install the required Python packages:

pip install -r requirements.txt

Start the backend server:

cd backend
python app.py

📌 Note:
Ensure that your merged and quantized LLM model (e.g., Qwen2.5-8B-Chat) is placed correctly, and the model path is properly configured in app.py.

By default, the Flask server will run at http://localhost:5000/.

---

3. Set Up the Frontend (React)

In a new terminal window, navigate to the frontend directory:

cd frontend
npm install

Start the frontend development server:

npm run dev

The React frontend will be available at http://localhost:5173/.

---

4. Prepare Knowledge Sources

Make sure the following knowledge files are placed under backend/data/:

• uk_study_guide.txt — Study abroad encyclopedia
• reconverted_prompt_completion.jsonl — Study abroad QA dataset

These files are preloaded automatically when the backend starts.

---

✅ 5. System is Ready!

• Access the system at http://localhost:5173/
• Start chatting or upload documents to enhance your questions with document-based knowledge.
• Enjoy the intelligent document-augmented conversation!

---

📚 Knowledge Sources

• User-Uploaded Documents (PDF/Word)
• Study Abroad Encyclopedia (uk_study_guide.txt)
• Study Abroad QA Dataset (reconverted_prompt_completion.jsonl)