Production RAG System with Multi-Agent Architecture

Portfolio Note: Portfolio recreation of production RAG platform built at Omfys Technologies.

🎯 Overview

Enterprise RAG platform processing 10K+ documents with LangChain, OpenAI GPT-4, PostgreSQL pgvector, FAISS achieving 90%+ relevance with <2s latency and 99.5% uptime.

📊 Key Metrics

• Documents: 10K+ (PDFs, Word, Confluence, Slack)
• Relevance: 90%+
• Latency: <2 seconds
• Uptime: 99.5%
• Hallucination Reduction: 65%

🛠️ Tech Stack

• LLM: OpenAI GPT-4, Anthropic Claude
• Framework: LangChain, LlamaIndex
• Vector DB: PostgreSQL pgvector, FAISS
• Orchestration: Docker, Kubernetes
• Monitoring: Prometheus, Grafana, Loki
• Load Balancing: Nginx
• Caching: Redis

⚡ Key Features

1. Multi-Agent Architecture

• Retrieval Agent: Document fetching with hybrid search
• Reformulation Agent: Query enhancement and expansion
• Synthesis Agent: Context-aware answer generation
• Attribution Agent: Source citation and verification

2. Hybrid Vector Search

• PostgreSQL pgvector (1536-dimensional embeddings)
• FAISS IVF index with product quantization
• Handles 50M+ chunks
• Multi-level caching with Redis

3. Production Infrastructure

• Docker orchestration with Kubernetes auto-scaling
• Nginx load balancing across 6 inference servers
• JWT authentication and tiered rate limiting
• Comprehensive monitoring with Prometheus/Grafana

4. Advanced RAG Techniques

• Recursive retrieval
• Query decomposition
• Re-ranking with cross-encoders
• Hallucination detection
• Source attribution

📁 Project Structure

rag-multiagent-system/
├── src/
│   ├── agents/              # Multi-agent system
│   ├── embeddings/          # Vector stores
│   ├── api/                 # FastAPI
│   └── monitoring/          # Prometheus metrics
├── kubernetes/              # K8s manifests
├── prometheus/              # Monitoring config
├── grafana/                 # Dashboards
└── README.md

🚀 Getting Started

git clone https://github.com/Amanroy666/rag-multiagent-system.git
cd rag-multiagent-system
pip install -r requirements.txt
docker-compose up -d

📈 Performance Results

Metric	Value
Answer Relevance	92%
Hallucination Rate	3.5% (65% reduction)
Average Latency	1.8s
Cache Hit Rate	78%
Uptime	99.5%

👤 Author

Aman Roy - Data Engineer at Omfys Technologies
📧 contactaman000@gmail.com | 💼 LinkedIn