AI / MLIn progress
Edge AI Lab — ML on the device
My next flagship: taking models from a Jupyter notebook down to resource-constrained hardware — quantisation, latency budgets and real sensor data. Combines both halves of my career.
In progress
PythonPyTorchTinyMLQuantizationRaspberry Pi
- 1Pick a sensor problem & collect data← now
- 2Train a baseline model in Python
- 3Quantise & prune for the edge
- 4Deploy & benchmark on device
- 5Write it up on the blog
▶ Try it · neural net playground
2 → 12 → 12 → 1click / drag to add points · shift = class Btap / drag to add points
epoch
0
loss
–
accuracy
–
points
0
A real neural network training live in your browser — no libraries, just math. Add points and watch it learn.
Why this project
Most ML portfolios stop at a notebook. Most embedded portfolios never touch a model. This project is where both meet: train a model, compress it, and make it run on real hardware within real latency and memory budgets.
Plan
- Pick a sensor-driven problem (e.g. audio keyword spotting or vibration anomaly detection).
- Train a baseline in Python and document the metrics.
- Quantise / prune it and measure accuracy vs. size vs. latency.
- Deploy on-device and write up everything on the blog.
This page will be updated as the project progresses.