Vibecode

• 

  Chat with Codebase as a Coder powered by Gemini

• 

  Chat with Codebase as an Advisor powered bu

• 

  Home Interface

Project Title:

VibeCode: Digital Twin Snapshots for AI-Native Codebase Management

Tagline:

Transform codebases into portable, AI-consumable PDFs with perfect restoration - powered by Gemini 3

Description: (Use your 200-word version)

[Paste your Gemini 3 integration description here]

How we built it:

Built with Python, PyQt6, and Gemini Flash 2.0. The system uses:

- Gemini Flash 2.0 for multi-agent orchestration (file selection, context generation, chat)
- Gemini Embedding 001 for semantic code search
- ChromaDB for persistent vector storage
- Model Context Protocol for external tool integration (Google Drive, GitHub)
- Custom PDF rendering with embedded JSON manifests for perfect restoration

The architecture loads entire codebases (~180K tokens) into Gemini's 1M+ context window,
enabling whole-program reasoning without chunking.

Challenges we ran into:

- Managing the 1M token context window efficiently with dynamic pruning
- Implementing perfect fidelity restoration from PDFs (SHA-256 checksums)
- Creating a multi-agent architecture where each agent specializes in different tasks
- Integrating MCP for seamless external tool access
- Handling Nigeria power outages during final submission! 😅

Accomplishments that we're proud of:

- Created a novel "Digital Twin" serialization format for codebases
- Achieved 100% restoration fidelity with cryptographic verification
- Built a production-ready GUI with PyQt6
- Implemented 4 specialized Gemini-powered agents
- Integrated external tools via Model Context Protocol
- Enabled Time Travel version comparison with AI-explained diffs

What we learned:

- How to maximize Gemini's large context window for whole-program analysis
- Multi-agent system design with specialized LLM roles
- Structured output handling with JSON mode
- PDF as a knowledge artifact format for AI systems
- Building agentic workflows with function calling

Built With

• chromadb
• gemini
• pyqt6
• python