embedded-ml

Machine Learning Systems: Foundations, Scaling, Agentic AI, and Physical AI (Vols I–IV) • Harvard CS249r | https://mlsysbook.ai

MINERVA - Minimal Inference Engine for Robust, Verifiable, and Authenticated ML. Encrypted, integrity-verified neural network inference for MCUs down to ATmega328P.

TinyML & Edge AI: On-device inference, model quantization, embedded ML, ultra-low-power AI for microcontrollers and IoT devices.

Pure-Rust prompt-injection detector with 1.5MB embedded MLP classifier. 98.40% accuracy, p50 14ms CPU inference, bindings for Python/JS/Go. Apache-2.0/MIT alternative to Rebuff (archived) and Lakera Guard.

Ahead-of-time compiler that fits ML models on embedded devices with hard memory budgets.

AIoT motion‑state detection pipeline: a simulated ESP32 + MPU6050 sensor (Velxio) streams accelerometer data over MQTT to Node‑RED, a TinyML model trained in Edge Impulse classifies Idle/Move/Shake on the edge, and predictions plus Shake alarms are visualized live on a ThingsBoard dashboard.

50.7x latency-optimized heterogeneous Vision Transformer (DeiT) hardware accelerator built on the Xilinx ZCU104 FPGA using DPU + custom HLS IPs.

Python ML for training a custom on-device cry model (knowledge-distilled from YAMNet, INT8, deployed on ESP32-S3)

Fajar Lang (fj) — Systems programming language for embedded ML & OS development. Compiler-enforced safety with @kernel/@device/@safe contexts. Rust-based compiler with Cranelift/LLVM backends. Made in Indonesia.

Suite of edge-AI tools and a web dashboard that help schools cut water, food, and energy waste by detecting leaks, sorting compost, and visualizing consumption in one conclusive dashboard.

ESP32 camera that escalates from gentle reminders to airhorn if you slouch

Hardware-aware face detection on Samsung GT-S7392 (ARM Cortex-A9)

Novel DBSCAN-based FPGA system for automatic modulation classification and NDA SNR estimation. O(n²)→O(n) complexity reduction. 71.7% lower power than state-of-the-art.

Generative horror adventure with a quantized transformer running locally in a Game Boy-compatible ROM.

Static INT8 AOT compiler that turns fixed PyTorch MCU models into standalone C11

Curated Edge AI resources for computer vision & audio: hardware, frameworks, benchmarks, literature, and communities (excluding mobile).

End-to-end TinyML pipeline: gesture recognition on Arduino Nano 33 BLE Sense — 1D CNN (97.6% acc, 26.9 KB INT8) + 5 ML baselines, BLE→WebSocket→web dashboard.

Notes and resources from Qualcomm On-device AI course, provided by DeepLearningAI

ChatTLM: a 10.9M-parameter language model that runs on a TI-Nspire CX II CAS graphing calculator, with the code, data and measurements behind the paper "Letting the Tools Do the Math"

Six-class activity classifier with compact softmax modeling, int8 quantized C++ inference, reproducible benchmarking, and an ESP32 replay sketch using UCI HAR data.