Projects / From idea to experiment
Smaller models.
Practical tools.
The goal is AI that can do useful work on-device. This page follows the build process, the tradeoffs, and the evidence needed to move from an idea to an open-source release.
An open path to on-device AI
Start with a focused task. Establish a baseline. Reduce the footprint, test the result, and publish enough detail for someone else to reproduce it.
Efficiency has to include usefulness. A smaller model is only progress when its quality and limitations are understood.
Proposed development roadmap
No dates committed- Planned
Choose a useful baseline
Define a narrow task, select an existing model, and measure quality, memory, and latency.
- Planned
Make it smaller
Compare quantization and compression approaches. Record what changes in accuracy and resource use.
- Planned
Test on actual devices
Evaluate local inference on a defined device. Measure runtime behavior and document failure cases.
- Planned
Share a reproducible release
Publish code, evaluation steps, and limitations. Review model and dataset licenses before distributing artifacts.
Other software work
Learning tools built for participation
A separate software project exploring live multiplayer learning. Future write-ups can show the interface, explain state management, and document the engineering decisions.
Explore the GitHub profile ↗Looking for data analysis and notebooks?
Visit the research journal ↗