AI Developer Kits

More Than a Model Zoo

Build and ship on-device AI quickly. Our ADKs help you go from train/test to demo/deploy with simple YAML and BYO tasks, datasets, and models—covering domains like audio, sleep, and heart monitoring.

heartKIT

soundKIT

Highlights

Pick a Kit, Ship Fast

Start with sleepKIT, heartKIT, or soundKIT—each includes domain tasks, datasets, model templates, and pre-trained demos. Go from idea to on-device prototype in hours.

BYO Anything

Switch tasks, models, datasets, and training settings easily with YAML. Use your own data and architectures without refactoring—your workflow, your rules.

Runtime, Your Way

Run on TFLM, heliaRT, or heliaAOT. Pick the path that fits your latency, footprint, and power budget.

Built for Battery Life

Deploy to Apollo ultra-low-power SoCs with optimized pipelines and web-based demos. Validate performance on real hardware—ideal for wearables, wellness, RPM, and smart audio.

What Is an ADK?

An AI Development Kit (ADK) is software, not a box. It’s a ready-to-run workspace that lets you train, test, demo, and deploy AI to Ambiq Apollo—fast.

Software toolkit : tasks, datasets, templates, web demos
YAML-driven workflow: data → features → train/test/demo/deploy
Not hardware: no dev board or sensors
Not a dump: more than models or scripts
Beats “model zoos”: end-to-end, device-ready (heliaAOT/heliaRT), battery-tuned

Available ADKs

Train, test, and deploy real-time cardiac models on Ambiq’s ultra-low power SoCs. Perfect for fitness wearables, wellness devices, and remote patient monitoring—covering denoise, ECG/PPG segmentation, rhythm, and beat classification.

Build on-device sleep analytics for wearables and remote monitoring on Ambiq’s ultra-low-power SoCs. sleepKIT includes Detect (sustained sleep/inactivity), Stage (light/deep/REM), and Apnea (apnea/hypopnea) with ready-to-train models and demos.

Build on-device voice AI for wearables, earbuds, and smart home devices on Ambiq’s ultra-low-power SoCs. soundKIT includes Speech Enhancement, Keyword Spotting (TFLite/TFLM-compatible), Voice Activity Detection (VAD), and Speaker Identification (ID) for fast, private, low-latency experiences.