A powerful Retrieval-Augmented Generation (RAG) system built with modern AI technologies, featuring hybrid search capabilities and Thai language support for health-related information.

• 🔍 Hybrid Search: Combines semantic search (vector) with keyword search (BM25) for optimal results
• 🧠 AI-Powered Q&A: Uses Large Language Models for intelligent question answering
• 🌐 Web Interface: User-friendly Streamlit interface with chat-like experience
• 🏥 Health Domain: Pre-loaded with Thai medical knowledge base
• ⚡ Real-time: Fast API backend with async processing
• 🐳 Containerized: Easy deployment with Docker
• 🔧 Extensible: Modular design for easy customization
• 💾 HuggingFace Embeddings: Uses BAAI/bge-m3 model for high-quality embeddings

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Streamlit UI  │◄──►│   FastAPI       │◄──►│   OpenSearch    │
│   (Frontend)    │    │   (Backend)     │    │   (Vector DB)   │
└─────────────────┘    └─────────────────┘    └─────────────────┘
         │                       │                       │
         │                       │                       │
         └───────────────────────┼───────────────────────┘
                                 │
                       ┌─────────▼─────────┐
                       │   HuggingFace     │
                       │  Embeddings +     │
                       │     Ollama LLM    │
                       └───────────────────┘

• Docker & Docker Compose
• Python 3.10+ (via Miniconda/Anaconda)
• Git
• PyTorch 2.6+ (Required for security fix with latest transformers)
• CUDA-compatible GPU (optional, for faster embeddings)

git clone https://github.com/amornpan/Generic-RAG.git
cd Generic-RAG

For Windows:

1. Download installer from: https://repo.anaconda.com/miniconda/Miniconda3-latest-Windows-x86_64.exe
2. Install and check "Add Miniconda3 to my PATH environment variable"
3. Open new Command Prompt or PowerShell

For Linux/macOS:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
source ~/.bashrc

# Accept ToS for the main Anaconda channels
conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/main
conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/r

# Create conda environment
conda create -n generic_rag_env python=3.10 -y
conda activate generic_rag_env

# if error
#conda tos accept --override-channels --channel https://repo.anaconda.com/pkgs/msys2

# conda init

# conda create -n generic_rag_env python=3.10 -c conda-forge -y

# Install PyTorch 2.6+ first (IMPORTANT: Required for security fix)
# For CPU version:
#pip install torch>=2.6.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

# For GPU version (CUDA 11.8):
# pip install torch>=2.6.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

# Install other dependencies
pip install -r requirements.txt

Make sure Docker Desktop is running on your system.

For Windows (PowerShell/Command Prompt):

docker run -d --name opensearch-node -p 9200:9200 -p 9600:9600 -e "discovery.type=single-node" -e "bootstrap.memory_lock=true" -e "OPENSEARCH_JAVA_OPTS=-Xms1g -Xmx1g" -e "DISABLE_INSTALL_DEMO_CONFIG=true" -e "DISABLE_SECURITY_PLUGIN=true" opensearchproject/opensearch:2.11.1

For Linux/macOS:

docker run -d --name opensearch-node -p 9200:9200 -p 9600:9600 -e "discovery.type=single-node" -e "bootstrap.memory_lock=true" -e "OPENSEARCH_JAVA_OPTS=-Xms1g -Xmx1g" -e "DISABLE_INSTALL_DEMO_CONFIG=true" -e "DISABLE_SECURITY_PLUGIN=true" opensearchproject/opensearch:2.11.1

Wait for OpenSearch to start (30-60 seconds), then:

For Windows (PowerShell):

Invoke-RestMethod -Uri "http://localhost:9200/_search/pipeline/hybrid-search-pipeline" `
  -Method PUT `
  -ContentType "application/json" `
  -Body '{
    "description": "Post processor for hybrid search",
    "phase_results_processors": [
      {
        "normalization-processor": {
          "normalization": {"technique": "min_max"},
          "combination": {
            "technique": "arithmetic_mean",
            "parameters": {"weights": [0.3, 0.7]}
          }
        }
      }
    ]
  }'

For Linux/macOS:

curl -X PUT "localhost:9200/_search/pipeline/hybrid-search-pipeline" \
  -H "Content-Type: application/json" \
  -d '{
    "description": "Post processor for hybrid search",
    "phase_results_processors": [
      {
        "normalization-processor": {
          "normalization": {"technique": "min_max"},
          "combination": {
            "technique": "arithmetic_mean",
            "parameters": {"weights": [0.3, 0.7]}
          }
        }
      }
    ]
  }'

1. Download and install Ollama from: https://ollama.ai/download
2. For mac/linux

   curl -fsSL https://ollama.com/install.sh | sh

3. Start Ollama service:

   ollama list

4. Pull required model:

   ollama pull qwen2.5:7b

# Create vector index with HuggingFace embeddings
python embedding.py

This will:

• Load markdown documents from md_corpus/ directory
• Create embeddings using BAAI/bge-m3 model
• Store vectors in OpenSearch
• Save index to md_index.pkl

# Terminal 1: Start API server
python api.py

# Terminal 2: Start Streamlit UI
streamlit run app.py

streamlit run app.py --server.address 0.0.0.0 --server.port 8501

• Web UI: http://localhost:8501
• API Docs: http://localhost:9000/docs
• OpenSearch: http://localhost:9200

The UI will show the status of all services (API, Ollama, OpenSearch) at the top.

Generic-RAG/
├── README.md                 # This file
├── requirements.txt          # Python dependencies
├── .env                     # Environment variables (optional)
│
├── embedding.py             # Data indexing with HuggingFace embeddings
├── api.py                   # FastAPI backend with HuggingFace embeddings
├── app.py                   # Streamlit frontend
│
├── md_corpus/               # Knowledge base (Markdown files)
│   ├── 1.md                # German measles (หัดเยอรมัน)
│   ├── 2.md                # Cholera (อหิวาตกโรค)
│   ├── 44.md               # Cataract (ต้อกระจก)
│   └── 5555.md             # GERD (กรดไหลย้อน)
│
└── md_index.pkl            # Saved index (created after running embedding.py)

Create a .env file (optional) to override defaults:

# OpenSearch Configuration
OPENSEARCH_ENDPOINT=http://localhost:9200
OPENSEARCH_INDEX=dg_md_index

# API Configuration
API_HOST=0.0.0.0
API_PORT=9000

The system includes Thai medical information covering:

1. หัดเยอรมัน (German Measles/Rubella) - Symptoms, causes, treatment
2. อหิวาตกโรค (Cholera) - Bacterial infection causing severe diarrhea
3. ต้อกระจก (Cataract) - Eye condition common in elderly
4. กรดไหลย้อน (GERD) - Gastroesophageal reflux disease

1. Place markdown files in md_corpus/ directory
2. Delete old index:

   curl -X DELETE "localhost:9200/dg_md_index"

3. Run python embedding.py to reindex
4. Restart the API server

embedding_model_name = 'BAAI/bge-m3'  # Current model
# Can change to other HuggingFace models like:
# embedding_model_name = 'intfloat/multilingual-e5-large'

llm_model = "qwen2.5:7b"  # Current model
# Can change to:
# llm_model = "qwen2.5:7b"  # Better quality
# llm_model = "llama2:13b"  # Alternative

Remember to pull new Ollama models:

ollama pull qwen2.5:7b

# Test API health
curl http://localhost:9000/health

# Test search endpoint
curl -X POST "http://localhost:9000/search" \
  -H "Content-Type: application/json" \
  -d '{"query": "อาการของโรคหัดเยอรมัน"}'

# View all indices
curl -X GET "localhost:9200/_cat/indices?v"

# View index details
curl -X GET "localhost:9200/dg_md_index?pretty"

# Count documents in index
curl -X GET "localhost:9200/dg_md_index/_count?pretty"

# View sample documents
curl -X GET "localhost:9200/dg_md_index/_search?pretty" -H 'Content-Type: application/json' -d'
{
  "size": 5,
  "query": {
    "match_all": {}
  }
}'

1. "No module named 'xxx'" Error

   pip install -r requirements.txt

2. PyTorch Security Error (torch.load vulnerability)

   # This error occurs with transformers 4.37+ and PyTorch < 2.6
   # Solution: Upgrade PyTorch to 2.6+
   pip install torch>=2.6.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
   
   # Alternative: Downgrade transformers
   # pip install transformers==4.36.0 tokenizers==0.15.0

3. OpenSearch connection failed

   # Check if OpenSearch is running
   docker ps
   # Check OpenSearch health
   curl http://localhost:9200/_cluster/health

4. Ollama not responding

   # Check installed models
   ollama list

5. No search results

   # Check document count
   curl -X GET "localhost:9200/dg_md_index/_count"
   # If 0, rerun embedding.py
   python embedding.py

6. GPU not detected

   # Check PyTorch GPU support
   python -c "import torch; print(torch.cuda.is_available())"
   # If False, install GPU version of PyTorch
   pip install torch>=2.6.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

1. Use GPU: Significantly faster for embeddings
2. Batch Processing: embedding.py processes in batches automatically
3. Adjust Chunk Size: In embedding.py, modify chunk_size=1024
4. Use Larger LLM: For better answers, use qwen2.5:7b or larger

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Test thoroughly
5. Submit a pull request

This project is licensed under the MIT License.

• LlamaIndex - RAG framework
• OpenSearch - Vector database
• Ollama - Local LLM runtime
• HuggingFace - Embedding models
• Streamlit - Web interface
• FastAPI - API framework

Made with ❤️ for the AI community