Building Modular Python Components for Multimodal Data Infrastructure

The rapid growth of multimodal AI has created an urgent need for flexible, efficient, and scalable data infrastructure. Handling diverse modalities, from images and text to signals, tabular data, and structural information, requires modular tools that can support seamless loading, preprocessing, and integration. Yet, building such infrastructure remains a challenge, especially when dealing with missing or heterogeneous data.

This mini-hackathon focuses on designing and prototyping standardised, flexible, and scalable PyTorch-based datasets and dataloaders. The solutions will potentially contribute to the future multimodal data infrastructure (see the Open Multimodal AI Benchmark funding call for more details) our UK Open Multimodal AI Network (UKOMAIN) community aims to build. A starter codebase with I/O functions and example datasets will be provided for multiple modalities, including images, text, signals, and tabular data. You are also welcome to incorporate additional data sources if you wish.

In just 4 hours, teams will explore solutions that are scalable, extensible, and generalisable, helping to power the next generation of multimodal learning.

This mini-hackathon welcomes researchers and practitioners with basic Python programming experience. To participate fully, please ensure the following:

• Bring a laptop with Wi-Fi capability
• Have a GitHub account https://github.com/signup to make contributions and use GitHub Discussions.
• Please set up a working Python environment with PyTorch, PyTorch Lightning, and PyTorch Geometric (PyG) installed before the hackathon. Other smaller dependencies can be installed at the event. If you run into difficulties installing these packages, facilitators will be available on site to help you.

Mini-Hackathon Codebase

Data Examples

Hackathon Resources & Highlights

This mini-hackathon is open to all attendees of the Third Workshop on Multimodal AI.

The registration for the workshop has been closed. If you have registered for the workshop, you should have received an email with a form to register for the mini-hackathon.

Contact Us

Email the organisers: ukomain-mmai25@googlegroups.com