Digital Twin

Your Digital Twin for people to know about you before the conversation starts

Comment

Inspiration

As founders/builders, we are constantly bombarded with context switching—investor emails, hiring candidates, code reviews, and scheduling nightmares. We asked ourselves: "What if we could clone ourselves?"

We didn't just want a chatbot; we wanted a Digital Twin that thinks like us, has access to our real-world tools, and protects our time. Neuralink isn't ready yet, so we built FounderOS ( A Digital Twin ).

What it does

My Digital Twin is an intelligent AI agent that acts as a gatekeeper and operator for your digital life.

  • Persona mimicking: Powered by Featherless.ai (Llama 3.1), it adopts the founder's specific tone and decision-making style.
  • Real-time Code Audits: It can search your actual GitHub repositories (e.g., checking middleware.ts for auth logic) and answer technical questions with context.
  • Gatekeeper for Time: It connects to Google Calendar to check real-time availability and negotiate meeting slots without you opening a tab.
  • Company Verification: It vets incoming leads/companies to see if they are legitimate (e.g., funding stage, location).
  • Emergency Line: If something is truly urgent, it bypasses the chat and sends a high-priority alert to your phone via Discord.

How we built it

We built a robust, decoupled 3-tier architecture:

  1. The Interface (Frontend):
    • built with React & Tailwind CSS.
    • Features a custom Audio Visualizer that reacts to the AI's "voice" intensity.
    • Cyberpunk/Terminal aesthetic to feel like a "command center."
  2. The Brain (Backend):
    • FastAPI (Python) server that manages the WebSocket connection.
    • Integrates with Featherless.ai for low-latency inference.
    • Acts as the MCP Client, translating user intent into tool execution commands.
  3. The Hands (MCP Server):
    • Built with Node.js, Hono, and the LeanMCP SDK.
    • We moved away from simple mocks to Real API Integrations:
      • Octokit for GitHub.
      • Google APIs for Calendar.
      • Discord Webhooks for alerts.
    • Exposes a standard Model Context Protocol (MCP) endpoint over HTTP/SSE.

Challenges we ran into

  • The MCP Learning Curve: Implementing the Model Context Protocol from scratch was complex. We initially faced stability issues with the raw SDK.
  • Server Communication: We migrated from a standard I/O subprocess model to a robust HTTP/SSE architecture (using Hono) to allow the MCP server to run independently and scale.
  • Real-Time Auth: configuring the Google Service Account to interact seamlessly with a personal calendar required some deep dives into GCP IAM permissions.

Accomplishments that we're proud of

  • True Agentry: It's not just RAG; the AI decides when to use a tool. If you ask "How's my schedule?", it calls the calendar. If you say "Alert me", it calls Discord.
  • LeanMCP Integration: We successfully refactored our code to use the modern @leanmcp/core decorators, making our tool definitions clean and type-safe.
  • The Vibe: We spent time polishing the "Neural Link" aesthetic. The boot-up sequences, log streams, and visualizers make it feel like a piece of sci-fi tech.

What we learned

  • MCP is powerful: Decoupling the "Brain" (LLM) from the "Tools" (MCP Server) is a game-changer. It allows us to swap out models (Llama, GPT-4, etc.) without breaking our tool integrations.
  • Context is King: Giving the AI access to the file system and live APIs transforms it from a "search engine" to a "co-worker."

What's next for Digital Twin

  • Voice Mode: Adding Speech-to-Text and Text-to-Speech for a full "Iron Man / JARVIS" experience. Kept on hold due to ElevenLabs credit unavailability.
  • Autonomous Mode: Allowing the twin to run in the background, proactively checking emails and drafting replies without user input.
  • More Integrations: Linear (for issue tracking) and Slack (for team comms) are next on the list.

Built With

Share this project:

Updates