Machine Learning Systems

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

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

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.

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

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

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

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.

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

This is open source library for creating artificial neural network in c programming language for general purpose use.

Estudo comparativo de arquiteturas de deep learning (CNN 1D, MLP, GRU, LSTM) para predição de temperatura em sistemas TinyML. Análise de performance, precisão e viabilidade para deploy em RP2040 com fusão de sensores AHT20/BMP280. Horizontes de 5, 10 e 15 minutos.

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

CS2 Skin Preview & Customization Utility for Weapons and Inventory is a visual tool for exploring and customizing weapon and inventory appearances in Counter-Strike 2, designed for previews, loadout styling, and cosmetic experimentation.

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.

Real-time motor speed classification using TinyML on Raspberry Pi Pico W. MLP neural network trained with TensorFlow deployed on embedded hardware (5.3 KB model). Classifies motor vibration into 4 speed levels using MPU6050 accelerometer with live OLED display feedback. Complete ML workflow from data collection to edge deployment.

Deploy and manage ML models at the edge — OPC-UA integration, PLC connectivity, real-time inference on embedded hardware for sub-millisecond decisions

LiDAR-based object classification system using CNN for autonomous robot navigation. Achieved 97% accuracy in classifying household objects with real-time operation. Developed at Imperial College London.

Hands-on labs for ML engineers to deploy edge AI inference with ExecuTorch on Arm platforms using PyTorch, XNNPACK, and Ethos-U (educational)

Don't Think It Twice: Exploit Shift Invariance for Efficient Online Streaming Inference of CNNs

Multiposition heart sound analysis