# Edge AI Tinyml

> Use when deploying ML models to edge devices.

- Skill: `loopyluci/edge-ai-tinyml` (Agent Skill)
- Install (CLI): `npx skillmds@latest add loopyluci/edge-ai-tinyml`
- Raw SKILL.md: https://api.skillmd.com/api/skills/loopyluci/edge-ai-tinyml/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- License: MIT
- Author: LoopyLuci (https://skillmd.com/u/loopyluci)
- Updated: 2026-09-08
- Page: https://skillmd.com/skills/loopyluci/edge-ai-tinyml

---


# Edge AI and TinyML

Deploying ML models to edge devices — from model optimization (quantization, pruning) through TensorFlow Lite Micro, ONNX Runtime, and deployment on MCUs and edge hardware.

## When to Use

- Running ML models on resource-constrained devices
- Reducing cloud dependency and latency
- Privacy-preserving on-device inference
- IoT sensor data processing at the edge

## Edge ML Pipeline

```python
class EdgeMLOptimizer:
    """Optimize models for edge deployment."""
    
    @staticmethod
    def quantize_to_int8(model_path: str, output_path: str, 
                          representative_dataset) -> str:
        import tensorflow as tf
        converter = tf.lite.TFLiteConverter.from_saved_model(model_path)
        converter.optimizations = [tf.lite.Optimize.DEFAULT]
        converter.representative_dataset = representative_dataset
        converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
        tflite_model = converter.convert()
        
        with open(output_path, 'wb') as f:
            f.write(tflite_model)
        return output_path
    
    @staticmethod
    def estimate_memory(model_path: str) -> Dict:
        import os
        size = os.path.getsize(model_path)
        return {'model_size_bytes': size, 'model_size_kb': round(size / 1024, 1)}
```

## Verification Checklist

- [ ] Quantization method chosen (INT8, FP16) for target hardware
- [ ] Model fits within edge device memory (RAM + flash)
- [ ] Inference latency acceptable for use case
- [ ] Accuracy validated post-quantization
- [ ] Hardware support verified (TFLite Micro, ONNX Runtime, or vendor SDK)
- [ ] Power consumption measured (battery-powered devices)
- [ ] OTA update mechanism for model updates