RAG Chatbot - FIT HCMUS

Chatbot tu van hoc vu cho sinh vien Khoa Cong Nghe Thong Tin (FIT), truong Dai hoc Khoa Hoc Tu Nhien - DHQG TP.HCM. Su dung kien truc RAG (Retrieval-Augmented Generation) voi ReAct Agent, Memory, va Vector Database.

System Design

graph TB
    subgraph Frontend["FRONTEND (React + Vite)"]
        ChatBot["ChatBot.jsx"]
        HomePage["HomePage.jsx"]
        FAQPage["FAQPage.jsx"]
        IssuePage["IssuePage.jsx"]
    end

    ChatBot -->|"GET /rag/{source}?q=...&mode=...&session_id=..."| API

    subgraph Backend["BACKEND (FastAPI)"]
        API["routes.py<br/><small>Rate Limiting (slowapi)</small>"]
        RAGSvc["RAGService<br/><small>rag_service.py</small>"]
        API --> RAGSvc

        subgraph ClassicPipe["Classic Pipeline"]
            Retriever["RetrieverManager<br/><small>retriever.py</small>"]
            Reranker["Reranker<br/><small>reranker.py</small>"]
            Generator["Generator<br/><small>generator.py</small>"]
            Retriever --> Reranker --> Generator
        end

        subgraph AgenticPipe["Agentic Pipeline"]
            Agent["ReactRAGAgent<br/><small>agent.py (LangGraph)</small>"]
            Tools["Tools<br/><small>qdrant_search | fit_website_search</small>"]
            Memory["MongoSessionMemoryStore<br/><small>memory.py</small>"]
            Agent --> Tools
            Agent --> Memory
        end

        RAGSvc -->|"mode=classic"| Retriever
        RAGSvc -->|"mode=agentic"| Agent

        History["ChatHistoryStore<br/><small>SQLite</small>"]
        Metrics["Prometheus<br/><small>/metrics</small>"]
    end

    subgraph External["EXTERNAL SERVICES"]
        Qdrant["Qdrant<br/>Vector DB"]
        Gemini["Google Gemini<br/>LLM API"]
        HF["HuggingFace<br/>Embeddings"]
        MongoDB["MongoDB<br/>Session Memory"]
    end

    Retriever --> Qdrant
    Tools --> Qdrant
    Generator --> Gemini
    Agent --> Gemini
    Retriever --> HF
    Memory --> MongoDB

    subgraph Monitoring["MONITORING (Docker)"]
        Prom["Prometheus :9090"]
        Graf["Grafana :3000"]
        Prom --> Graf
    end

    Metrics --> Prom

    subgraph Pipeline["DATA PIPELINE"]
        cralw["crawl_fit_pdfs.py<br/><small>FIT PDF crawler</small>"]
        ocr["llm_ocr_pdf.py<br/><small>OCR - legacy</small>"]
        TXT["Database/*.txt"]
        Loader["loaders.py"]
        Splitter["splitters.py"]
        Embed["embeddings.py"]
        cralw-->ocr-->TXT --> Loader --> Splitter --> Embed --> Qdrant
    end

Loading

Project Structure

RAG_Chatbot/
├── AGENTS.md
├── CLAUDE.md
├── LICENSE
├── readme.md
├── back-end/
│   ├── main.py
│   ├── requirements.txt
│   ├── app/
│   │   ├── api/routes.py
│   │   ├── config/
│   │   │   ├── config.py
│   │   │   └── prompts.py
│   │   ├── core/
│   │   │   ├── dependencies.py
│   │   │   └── memory_agent.py
│   │   ├── rag/
│   │   │   ├── agent.py
│   │   │   ├── retriever.py
│   │   │   ├── reranker.py
│   │   │   ├── generator.py
│   │   │   ├── llm.py
│   │   │   └── tools.py
│   │   ├── services/rag_service.py
│   │   └── storage/
│   │       ├── history.py
│   │       └── memory.py
│   └── tests/
├── front-end/
│   ├── package.json
│   ├── src/
│   │   ├── main.jsx
│   │   ├── components/
│   │   ├── pages/
│   │   └── services/
├── Data/
│   ├── run_pipeline.py
│   ├── requirements.txt
│   ├── Database/
│   └── pipeline/
├── docs/
├── eval/
└── monitoring/

Query Parameters for /rag/{source}:

• q (required) — Query text
• mode — classic or agentic (default: classic)
• session_id — Session ID for memory continuity (agentic mode)
• debug — Include thought_process in response

---

Cai dat

Backend

cd back-end
cp .env.example .env     # Dien GOOGLE_API_KEY, QDRANT_URL, QDRANT_API_KEY
pip install -r requirements.txt
python main.py           # http://127.0.0.1:8000

Frontend

cd front-end
cp .env.example .env     # Dien VITE_API_BASE_URL
npm install
npm run dev              # http://localhost:5173

Data Pipeline

cd Data
cp .env.example .env     # Dien QDRANT_URL, QDRANT_API_KEY, HUGGINGFACE_API_KEY
python run_pipeline.py   # Build vector index

Monitoring

docker compose -f monitoring/docker-compose.monitoring.yml up -d
# Prometheus: http://localhost:9090
# Grafana: http://localhost:3000 (admin/admin)

Tests

cd back-end
pip install pytest
python -m pytest tests/ -v    # 78 test cases

Demo

YouTube Demo

---

Contributing

1. Fork repo
2. Tao branch moi (git checkout -b feature/ten-feature)
3. Commit thay doi (git commit -m "feat: mo ta thay doi")
4. Push branch (git push origin feature/ten-feature)
5. Tao Pull Request

Vui long dam bao chay tests truoc khi tao PR:

cd back-end && python -m pytest tests/ -v

Tai lieu tham khao (References)

Papers

Paper	Tac gia	Mo ta
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks	Lewis et al., 2020	Kien truc RAG goc — ket hop retrieval voi generation
ReAct: Synergizing Reasoning and Acting in Language Models	Yao et al., 2022	ReAct Agent — vong lap Think-Act-Observe
Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks	Reimers & Gurevych, 2019	Sentence embeddings cho semantic search

Documentation

Cong nghe	Tai lieu
LangChain	python.langchain.com/docs
LangGraph	langchain-ai.github.io/langgraph
FastAPI	fastapi.tiangolo.com
Qdrant	qdrant.tech/documentation
Google Gemini API	ai.google.dev/docs
Sentence-Transformers	sbert.net
React	react.dev
Vite	vite.dev/guide
TailwindCSS	tailwindcss.com/docs
DaisyUI	daisyui.com
Prometheus	prometheus.io/docs
Grafana	grafana.com/docs

Acknowledgments

• Khoa Cong nghe Thong tin (FIT) — Truong Dai hoc Khoa Hoc Tu Nhien, DHQG TP.HCM
• Cac du lieu huan luyen duoc thu thap tu fit.hcmus.edu.vn

License

Du an nay duoc phat hanh theo MIT License.

MIT License - Copyright (c) 2023 phatjk