Skip to content

KathiraveluLab/Mojito

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Mojito

Mojo-based Integrated Task Orchestrator

"Execution comes to you. Data does not leave you."

Mojito is a high-performance, privacy-preserving federated orchestration framework built with the Mojo programming language. It is designed to enable complex workflows, such as transfer learning across sensitive data silos (Healthcare, Finance, etc.), while maintaining strict GDPR compliance and "near-zero" data movement.


Key Philosophy

Traditional distributed systems move massive datasets to centralized compute clusters. Mojito flips this paradigm:

  1. Data Stays Home: Raw data never leaves the local environment.
  2. Lightweight Execution: Tiny, Mojo-native binaries (<50MB) are pushed to the data.
  3. Privacy-First: Only anonymized parameter deltas (gradients) are returned for aggregation.
  4. Hardware-Native: Leverages Mojo's SIMD and ownership model for C-level performance on local hardware.

Installation

Mojo is required to run the local training loops. The recommended way to install Mojo on Linux is via pixi:

  1. Install Pixi:
    curl -fsSL https://pixi.sh/install.sh | sh
  2. Add Mojo to your project:
    pixi add mojo
  3. Enter the environment:
    pixi shell

For detailed installation options and OS-specific instructions, refer to the official Modular documentation.


Technical Spike: Local Training Loop

The project recently reached a milestone with the implementation of a pure Mojo local training loop:

  • File: mojito_local_training.mojo
  • Features:
    • SIMD-Optimized: Direct hardware mapping for weight updates.
    • No Python Overhead: 100% Mojo-native code for minimal container footprints.
    • Differential Privacy Ready: Structured to support noise-injection for gradient protection.

Running the Spike

If you have Mojo installed, you can run the local training simulation directly:

mojo run mojito_local_training.mojo

To build a standalone binary for containerized deployment:

mojo build mojito_local_training.mojo


Compliance

  • GDPR: Built with data minimization and purpose limitation at the core.
  • HIPAA/PCI-DSS: Designed to eliminate the primary audit liability of data transit.

Documentation

For more details on the design discussions and technical specifications, refer to the conversation logs and agent artifacts in the project framework.

About

Mojo-based Integrated Task Orchestrator

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages