Skip to content
GenAI Realtime Assistant
Feb 2025 - May 2025

GenAI Realtime Assistant

Multi-criterial AI assistant with GraphRAG and VectorDB

01. The Challenge

Traditional chatbots lack contextual understanding and struggle with complex, multi-faceted queries requiring knowledge graph reasoning.

02. The Solution

Built a real-time GenAI assistant using GraphRAG for knowledge graph integration, LangChain for orchestration, and Neo4j for graph storage.


Overview

The GenAI Realtime Assistant represents a leap forward in how we interact with knowledge bases. By combining GraphRAG with vector databases, we achieved a system that understands not just the "what" but the "how" and "why" of complex queries.


Key Challenges

  • **Latency:** Traditional RAG pipelines were too slow for real-time conversation.
  • **Context Loss:** Long conversations often lost context.
  • **Hallucination:** Pure LLM approaches were prone to making things up.

    • The Solution

    We implemented a multi-agent system using LangGraph where specialized agents handle different aspects of the query. The Knowledge Graph (Neo4j) provides structured ground truth, while the Vector DB (FAISS) handles unstructured semantic search.


    Architecture Highlights

  • **Router Agent:** Classifies intent and routes to the best sub-agent.
  • **Graph Agent:** Traverses Neo4j to find connected entities.
  • **Vector Agent:** Retrieves semantic matches from documentation.
  • **Synthesis Agent:** Combines all retrieval results into a coherent answer.

    • System Architecture

    LangGraph for multi-agent workflows
    Neo4j for knowledge graph storage
    FAISS for vector similarity search
    OpenAI GPT-4 for generation
    FastAPI backend with WebSocket support

    Project Links

    Technologies

    LangChainLangGraphOpenAINeo4jFAISSFastAPIPython

    Key Impact

    Real-time contextual responses
    Multi-criterial query handling
    Knowledge graph reasoning