Edge & On-Device AI

Shrink the model, keep the magic

Quantization, NPUs, mobile runtimes, and federated learning — shipping real intelligence where the network can't reach and the battery won't wait.

Advanced4 modules · 12 lessons~118 min1180 XPAligned with Google Coral/TensorFlow Lite + TinyML literatureStart with: Deep Learning

Progress0/12

Module

Why Compute Moves to the Edge

The physics and economics of on-device inference, the memory wall, and the hardware landscape

The Case for On-Device8 min · 80 XPThe Case for On-Device

The Memory Wall10 min · 100 XPThe Memory Wall

The Hardware Landscape10 min · 100 XPThe Hardware Landscape

Module

Shrinking the Model

Quantization mechanics, sub-8-bit methods, pruning, distillation, and taming the KV cache

Quantization: The Mechanics11 min · 100 XPQuantization: The Mechanics

Extreme Quantization: GGUF & Friends11 min · 110 XPExtreme Quantization: GGUF & Friends

Pruning & Distillation10 min · 100 XPPruning & Distillation

KV-Cache & Runtime Memory10 min · 100 XPKV-Cache & Runtime Memory

Module

Runtimes & Deployment

The software layer that reaches the silicon — and the pipeline that gets a model to production on-device

On-Device Runtimes10 min · 100 XPOn-Device Runtimes

The Deployment Pipeline9 min · 90 XPThe Deployment Pipeline

Module

Systems Reality

Battery and thermals as hard constraints, federated learning's real shape, and hybrid architectures

Battery, Thermals & Scheduling9 min · 90 XPBattery, Thermals & Scheduling

Federated Learning on the Edge10 min · 100 XPFederated Learning on the Edge

Hybrid Architectures10 min · 110 XPHybrid Architectures

Capstone project

Shrink the model

Take a model and make it small — quantization, pruning, or distillation. Measure the size/accuracy trade-off and find the sweet spot for a device you own.

Deliverable: A notebook comparing original vs. compressed model on size, speed, and accuracy.

Milestones

Finish every lesson to earn the Edge & On-Device AI Certificate.