ElasticMind AI — Context-Aware MCP - RAG Chatbot

An AI chatbot that converts uploaded files into smart, context-aware conversations using Gemini, Elasticsearch, and an MCP server — scalable for education, healthcare, and enterprises.

Comment

Inspiration

We wanted to build an intelligent assistant that combines Retrieval-Augmented Generation (RAG) with the Model Context Protocol (MCP) - creating a chatbot that doesn't just answer questions, but actually understands your documents through a modular, standardized architecture.

What it does

  • Upload PDFs, Word docs, Excel files, and images
  • Ask questions about your documents in natural language
  • Get AI-generated answers with relevant context from your files using RAG
  • Supports multi-user authentication (email, Google, phone)
  • Real-time chat with conversation history
  • MCP Server manages all document intelligence and AI operations

How we built it

We implemented a 3-tier RAG architecture with MCP Server:

  • Frontend (React): User interface for chat and document uploads
  • Backend (Node.js): Handles authentication and routes requests
  • MCP Server (Python): The RAG engine
    • Processes documents and extracts text
    • Generates embeddings using Sentence Transformers
    • Stores vectors in Elasticsearch for semantic search
    • RAG Pipeline: Retrieves relevant document chunks and augments Gemini AI prompts
    • Returns intelligent, source-backed answers

The MCP Server follows the Model Context Protocol, making our RAG system reusable and allowing other applications to connect to the same AI capabilities through a standardized API.

Challenges we ran into

  • Deployment headaches: Python 3.13 compatibility issues with PyTorch and scipy required downgrading to Python 3.11
  • Dependency conflicts: Resolving version mismatches between transformers, sentence-transformers, and huggingface-hub
  • Cross-service communication: Getting frontend, backend, and MCP server to work together across different domains
  • Large ML models: Optimizing PyTorch and embedding models for production (800MB+ dependencies)
  • CORS configuration: Managing authentication and requests across multiple deployed services

Accomplishments that we're proud of

  • Successfully built a production-ready RAG system with MCP integration
  • Enabled multi-user authentication with secure, real-time chat
  • Deployed complex ML pipelines across multiple platforms
  • Optimized embedding models for performance and scalability

What we learned

  • How to implement RAG (Retrieval-Augmented Generation) from scratch
  • Building production-ready MCP Server architecture
  • Managing complex multi-service deployments (Render + Vercel)
  • Handling large ML models in serverless environments
  • Firebase authentication integration with microservices

What's next for ElasticMind AI — Context-Aware MCP - RAG Chatbot

  • Add support for more document types (PowerPoint, CSV)
  • Implement streaming responses for faster user experience
  • Add document summarization and key insights extraction
  • Multi-language support for international documents

Built With

Share this project:

Updates