🚀 Advanced RAG Pipeline with Feast Feature Store + Milvus Vector DB

A production-ready Retrieval-Augmented Generation (RAG) pipeline with advanced feature store capabilities. Built with Feast feature store, Milvus-lite vector database, and Ollama LLM for document processing and intelligent question answering.

📸 Web Interface

Main Dashboard

Main dashboard showing query results with context-aware responses, source citations, and relevance scoring

Document Processing Pipeline

Document processing workflow showing upload progress, chunking process, and embedding generation

System Dashboard Statistics & Monitoring

System statistics dashboard displaying real-time metrics, document counts, and performance indicators

QnA

Question and Answer interface with query input, response display, and document source references

Query Interface

Primary query interface with smart question processing and real-time response generation

🏗️ System Architecture

                    🌐 Web Interface (localhost:8000)
                                    │
                    ┌───────────────▼───────────────┐
                    │        FastAPI Server        │
                    │     (Feast RAG Pipeline)     │
                    └───────────────┬───────────────┘
                                    │
        ┌───────────────────────────┼───────────────────────────┐
        │                           │                           │
        ▼                           ▼                           ▼
┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│  🤖 Ollama LLM  │    │ 🧠 Feast Store  │     │ 🗄️ Milvus-Lite  │
│                 │    │                 │    │                 │
│ • llama3.2:3b   │    │ • Feature Mgmt  │    │ • Vector Store  │
│ • Embeddings    │    │ • Online Store  │    │ • Similarity    │
│ • Generation    │    │ • Registry      │    │ • Collections   │
│ Port: 11434     │    │ • Milvus Backend│    │  File-based DB  │
└─────────────────┘    └─────────────────┘    └─────────────────┘

🔄 Data Flow

📄 Document Upload
       │
       ▼
┌─────────────────────────────────────────────────────────────┐
│                Document Processing Pipeline                 │
│                                                             │
│ 1. Parse Document → 2. Chunk Text → 3. Generate Embeddings │
│ 4. Store in Feast → 5. Sync to Milvus → 6. Index Vectors   │
└─────────────────────────────────────────────────────────────┘
       │
       ▼
┌─────────────────────────────────────────────────────────────┐
│                    Query Processing                         │
│                                                             │
│ 1. User Question → 2. Query Embedding → 3. Vector Search    │
│ 4. Retrieve Context → 5. LLM Generation → 6. Return Answer  │
└─────────────────────────────────────────────────────────────┘

✨ Key Features

🎯 Core Capabilities

🏪 Enterprise Feature Store - Feast for advanced feature management & serving
🚀 High-Performance Vector DB - Milvus-lite for scalable similarity search
🤖 Advanced LLM - Ollama with llama3.2:3b (3B parameters)
🧠 Smart Embeddings - all-MiniLM-L6-v2 (384 dimensions)
🔒 100% Local Processing - No data leaves your machine
🌐 Modern Web UI - Responsive interface with real-time updates

📄 Document Management

Multi-format Support - PDF, Markdown, Text, and Word documents
Smart Chunking - Intelligent text segmentation with overlap
Original Filename Preservation - Maintains document identity
Real-time Processing - Live feedback during upload
Seamless Clear Operations - PyMilvus-based collection management

🔍 Query & Retrieval

Semantic Search - Advanced vector similarity retrieval
Context-aware Responses - LLM with retrieved document context
Source Attribution - Detailed citations with relevance scores
Flexible Context Limits - Configurable result count
Real-time Stats - Live document count and system metrics

🛠️ System Management

Refresh Stats - Real-time system status updates
Clear All Documents - Complete collection reset with PyMilvus
Health Monitoring - Comprehensive system health checks
Performance Metrics - Document count, chunk statistics
Error Handling - Graceful failure recovery

🛠️ Tech Stack

Component	Technology	Version	Purpose
API Framework	FastAPI	0.104.1+	REST API & Web UI
Feature Store	Feast	0.51.0+	Feature management & registry
Vector Database	Milvus-lite	2.3.0+	File-based vector storage
LLM Engine	Ollama	Latest	Local language model serving
Language Model	llama3.2:3b	3B params	Text generation & reasoning
Embedding Model	all-MiniLM-L6-v2	384 dims	Document & query embeddings
Container Engine	Podman/Docker	Latest	Optional containerization

📋 Prerequisites

Python 3.12+ (required)
Poetry (recommended) or pip (alternative)
Ollama (for LLM serving)
At least 8GB RAM (16GB recommended for optimal performance)
5GB+ disk space (for models and data)

🚀 Quick Start

1. Clone and Setup

Option A: Poetry (Recommended)

# Clone the repository
git clone <repo-url>
cd rag-project

# Install Poetry (if not already installed)
curl -sSL https://install.python-poetry.org | python3 -

# Install dependencies with Poetry
poetry install --with=test,lint

# Activate virtual environment
poetry shell

Option B: Traditional pip

# Clone the repository
git clone <repo-url>
cd rag-project

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install Python dependencies
pip install -r requirements.txt

2. Initialize Feast Feature Store

# Initialize Feast feature store
cd feast_feature_repo
feast apply
cd ..

3. Start Ollama and Pull Models

# Start Ollama (in a separate terminal)
ollama serve

# Pull required models
ollama pull llama3.2:3b

4. Start the RAG Pipeline

With Poetry:

# Development mode (auto-reload)
make dev
# or
poetry run uvicorn src.api:app --host 0.0.0.0 --port 8000 --reload

# Production mode
make run
# or
poetry run uvicorn src.api:app --host 0.0.0.0 --port 8000

With pip:

# Start the FastAPI server
uvicorn src.api:app --host 0.0.0.0 --port 8000

5. Access the Application

Web UI: http://localhost:8000
API Docs: http://localhost:8000/docs

🔧 API Reference

Health Check

curl -X GET "http://localhost:8000/health"

Response:

{
  "status": "healthy",
  "feast_store": "True",
  "milvus_connection": "False",
  "embedding_model": "True",
  "message": "Feast RAG pipeline is running with unified Milvus backend"
}

System Statistics

curl -X GET "http://localhost:8000/stats"

Response:

{
  "pipeline_status": "ready",
  "vector_store_stats": {
    "collection_name": "rag_document_embeddings",
    "document_count": 3,
    "chunk_count": 15,
    "backend": "feast_milvus_lite"
  },
  "embedding_model": "all-MiniLM-L6-v2",
  "llm_model": "llama3.2:3b"
}

Document Ingestion

curl -X POST "http://localhost:8000/ingest" \
  -H "accept: application/json" \
  -H "Content-Type: multipart/form-data" \
  -F "file=@sample_docs/sample_document.md"

Response:

{
  "message": "Successfully ingested sample_document.md with 5 chunks using feast_official",
  "chunks_created": 5,
  "source": "sample_document.md",
  "metadata": {
    "storage_method": "feast_official",
    "status": "success",
    "file_name": "sample_document.md",
    "document_id": "feast_sample_document.md_5"
  }
}

Query Documents

curl -X POST "http://localhost:8000/query" \
  -H "accept: application/json" \
  -H "Content-Type: application/json" \
  -d '{
    "question": "What are the key features of this system?",
    "context_limit": 5
  }'

Response:

{
  "answer": "Based on the provided documents, the key features include...",
  "sources": [
    {
      "text": "Feature store capabilities with Feast...",
      "metadata": {
        "document_title": "sample_document.md",
        "chunk_index": 0,
        "file_path": "/path/to/document.md"
      },
      "similarity_score": 0.92
    }
  ],
  "context_used": 3,
  "relevance_scores": [0.92, 0.87, 0.84]
}

List Documents

curl -X GET "http://localhost:8000/documents"

Response:

{
  "documents": [
    {
      "title": "sample_document.md",
      "chunks": 5,
      "status": "processed"
    }
  ],
  "total_count": 1,
  "backend": "feast_milvus"
}

Clear All Documents

curl -X DELETE "http://localhost:8000/documents"

Response:

{
  "status": "success",
  "message": "Successfully cleared all documents from Feast Milvus database",
  "backend": "feast_milvus"
}

🌐 Web Interface Features

The web interface provides:

📤 Document Upload - Drag & drop interface supporting PDF, MD, TXT, DOCX
🔍 Intelligent Query - Natural language questions with context-aware responses
📊 System Dashboard - Real-time monitoring and statistics
🗂️ Document Management - List, view, and clear uploaded documents
🔄 Refresh Stats - Live system status updates

📁 Project Structure

rag-project/
├── src/                              # Core application code
│   ├── api.py                       # FastAPI server & endpoints
│   ├── feast_rag_pipeline.py        # Main RAG pipeline with Feast
│   ├── feast_rag_retriever.py       # Feast-based document retrieval
│   └── __init__.py
├── feast_feature_repo/               # Feast feature store configuration
│   ├── feature_store.yaml           # Feast configuration
│   ├── feature_definitions.py       # Feature views & entities
│   └── data/                        # Feature store data (excluded from git)
├── static/                           # Web interface files
│   ├── index.html                   # Main web UI
│   ├── script.js                    # Frontend JavaScript
│   └── style.css                    # UI styling
├── sample_docs/                      # Example documents & screenshots
│   └── ui_screenshots/              # Web interface screenshots
├── requirements.txt                  # Python dependencies
├── requirements-dev.txt              # Development dependencies
└── README.md                        # This file

🔧 Configuration

Feast Feature Store (`feast_feature_repo/feature_store.yaml`)

project: rag
provider: local
registry: data/registry.db
online_store:
  type: milvus
  path: data/online_store.db
  vector_enabled: true
  embedding_dim: 384
  index_type: "FLAT"
  metric_type: "COSINE"
offline_store:
  type: file
entity_key_serialization_version: 3
auth:
  type: no_auth

Environment Variables

# Optional configuration
export FEAST_REPO_PATH="feast_feature_repo"
export OLLAMA_HOST="localhost"
export OLLAMA_PORT="11434"
export LLM_MODEL="llama3.2:3b"
export EMBEDDING_MODEL="all-MiniLM-L6-v2"

🔍 Troubleshooting

Common Issues

Feast repository not found
```
cd feast_feature_repo
feast apply
```

Ollama model not available

# Pull required models
ollama pull llama3.2:3b
ollama list  # Verify models are installed

Collection not found after clear

cd feast_feature_repo && feast apply
# Restart the server to pick up recreated collection
# The system automatically handles collection recreation

Port conflicts

# Use different port
uvicorn src.api:app --host 0.0.0.0 --port 8000

Debug Commands

# Check service status
curl http://localhost:8000/health
curl http://localhost:8000/stats

# Check Ollama models
curl http://localhost:11434/api/tags

# Verify Feast setup
cd feast_feature_repo
feast entities list
feast feature-views list

🚀 Performance Optimization

Resource Requirements

Minimum: 8GB RAM, 4 CPU cores, 5GB storage
Recommended: 16GB RAM, 8 CPU cores, 20GB storage
Optimal: 32GB RAM, 16 CPU cores, 50GB SSD

Model Selection

# For better quality (requires more resources)
ollama pull llama3.2:3b

# For faster performance (lower quality)
ollama pull llama3.2:1b

# Update model in configuration
# Edit src/feast_rag_pipeline.py, line with model_name

File-based Milvus-lite Benefits

✅ Simplified deployment: No external containers required
✅ Single file database: Everything in feast_feature_repo/data/online_store.db
✅ Production ready: Proven integration with Feast
✅ Portable: Easy to backup and version control
✅ Fast startup: No complex container orchestration

🚀 Deployment

🐳 Docker Compose (Recommended)

# Using the deploy directory
cd deploy
./run.sh

# Or manually with docker-compose
docker-compose up --build -d

☸️ Kubernetes

# Apply Kubernetes manifests
kubectl apply -f deploy/k8s-deployment.yaml

# Check deployment status
kubectl get pods -n feast-rag-pipeline

🔧 Environment Configuration

Set environment variables for customization:

export RAG_API_PORT=9000
export RAG_LLM_MODEL=llama3.2:7b
export RAG_DEBUG_MODE=true

📋 Production Checklist

Configure persistent volumes for Feast data
Set appropriate resource limits (CPU/Memory)
Configure Ollama models for your use case
Set up monitoring and logging
Configure backup for Milvus database

📖 Full deployment guide: deploy/README.md

🧪 Testing

Run the test suite to verify your setup:

With Poetry (Recommended):

# Run all tests with verbose output
make test
# or
poetry run pytest -v

# Run tests with coverage
make test-cov
# or
poetry run pytest --cov=src --cov-report=html

# Run specific test class
poetry run pytest tests/test_rag_pipeline.py::TestFeastRAGPipeline -v

# Stop on first failure
poetry run pytest -x

# Format code before testing
make format

# Run all linting checks
make lint

With pip:

# Run all tests with verbose output
python -m pytest tests/test_rag_pipeline.py -v

# Run tests with coverage (install pytest-cov first)
pip install pytest-cov
python -m pytest tests/test_rag_pipeline.py --cov=src

# Run specific test class
python -m pytest tests/test_rag_pipeline.py::TestFeastRAGPipeline -v

# Stop on first failure
python -m pytest tests/test_rag_pipeline.py -x

Test Coverage:

✅ Pipeline initialization & error handling
✅ Document processing with Feast integration
✅ Query processing and retrieval
✅ Collection clearing operations
✅ Embedding generation functionality

🤝 Contributing

Fork the repository
Create a feature branch (git checkout -b feature/amazing-feature)
Commit your changes (git commit -m 'Add amazing feature')
Push to the branch (git push origin feature/amazing-feature)
Open a Pull Request

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

Feast - Feature store for ML
Milvus - Vector database for AI
Ollama - Local LLM serving
FastAPI - Modern Python web framework
Sentence Transformers - Embedding models

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
.github		.github
config		config
deploy		deploy
feast_feature_repo		feast_feature_repo
sample_docs		sample_docs
scripts		scripts
src		src
static		static
tests		tests
.gitignore		.gitignore
.pre-commit-config.yaml		.pre-commit-config.yaml
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
Makefile		Makefile
POETRY_MIGRATION.md		POETRY_MIGRATION.md
README.md		README.md
poetry.lock		poetry.lock
pyproject.toml		pyproject.toml
requirements.txt		requirements.txt
setup.cfg		setup.cfg

Folders and files

Latest commit

History

Repository files navigation

🚀 Advanced RAG Pipeline with Feast Feature Store + Milvus Vector DB

📸 Web Interface

Main Dashboard

Document Processing Pipeline

System Dashboard Statistics & Monitoring

QnA

Query Interface

🏗️ System Architecture

🔄 Data Flow

✨ Key Features

🎯 Core Capabilities

📄 Document Management

🔍 Query & Retrieval

🛠️ System Management

🛠️ Tech Stack

📋 Prerequisites

🚀 Quick Start

1. Clone and Setup

Option A: Poetry (Recommended)

Option B: Traditional pip

2. Initialize Feast Feature Store

3. Start Ollama and Pull Models

4. Start the RAG Pipeline

With Poetry:

With pip:

5. Access the Application

🔧 API Reference

Health Check

System Statistics

Document Ingestion

Query Documents

List Documents

Clear All Documents

🌐 Web Interface Features

📁 Project Structure

🔧 Configuration

Feast Feature Store (feast_feature_repo/feature_store.yaml)

Environment Variables

🔍 Troubleshooting

Common Issues

Debug Commands

🚀 Performance Optimization

Resource Requirements

Model Selection

File-based Milvus-lite Benefits

🚀 Deployment

🐳 Docker Compose (Recommended)

☸️ Kubernetes

🔧 Environment Configuration

📋 Production Checklist

🧪 Testing

With Poetry (Recommended):

With pip:

🤝 Contributing

📄 License

🙏 Acknowledgments

About

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Feast Feature Store (`feast_feature_repo/feature_store.yaml`)

Packages