# IOT x AI

## Pipeline

0. Customer provide the needs
1. Survey the appropriate structures and open source models to evaluate it
2. Finetune the model with specific domain dataset
3. POC stage: Simulate the model at Raspberri Pi
- Provide a simple GUI for demo
- Evaluate the metric/energy comsumption/ memory/computation
4. Optimize the model size/ computation
5. Porting the model at edge platform
6. Post evaluation the model
- Metrics
- Energy comsumption
- Memory Usage
- Computation comsumption

## Possible AI Models For Edge Computing

We could leverage the open source eco-system to construct the demo projects

- Qualcomm AI hub: https://aihub.qualcomm.com/
- Siliconlab models: https://siliconlabs.github.io/mltk/docs/python_api/models/examples/index.html
- Arm model zoo: https://github.com/ARM-software/ML-zoo
- mit-han-lab: mcu net: https://github.com/mit-han-lab/mcunet
- On-Device Training Under 256KB Memory: https://github.com/mit-han-lab/tiny-training
- Tiny whisper at Raspberri PI: https://github.com/ggerganov/whisper.cpp/discussions/166