adelmgadmi0/EdgeAI-STM32-Intent-Recognition

TinyML-based command intent recognition running on an STM32F446RE microcontroller. This project demonstrates how natural language commands (English + Tunisian dialect) can be classified on-device using TensorFlow Lite for Microcontrollers — without cloud processing. # EdgeAI-STM32-Intent-Recognition TinyML-based command intent recognition running on an STM32F446RE microcontroller. This project demonstrates how natural language commands (English + Tunisian dialect) can be classified on-device using TensorFlow Lite for Microcontrollers — without cloud processing. # STM32 TinyML Intent Recognition > A TinyML-based UART command classifier running on an STM32F446RE. This project demonstrates how a small neural network can classify text commands (English + Tunisian dialect) directly on a microcontroller using **STM32Cube.AI**. ## 🚀 Overview Instead of relying on traditional string matching (e.g., `strcmp`), this project leverages a trained neural network to classify incoming text commands into predefined intents. The model is deployed as Edge AI, meaning it runs entirely on the microcontroller with **zero cloud processing or internet connectivity required**. ## 📦 Features * **Runs Fully Offline (Edge AI):** All processing happens locally on the MCU. * **Bilingual Support:** Understands intents in both English and Tunisian dialect. * **Highly Optimized:** Uses X-CUBE-AI to generate memory-efficient C code tailored for STM32, minimizing RAM/Flash usage. * **No External Sensors Required:** Operates strictly via UART text streams. ## 🎯 Supported Intents | Intent | Expected Action | Example Triggers (English & Tunisian) | | :--- | :--- | :--- | | **`LED_ON`** | Turns the LED ON | *turn on led, activate, cha3el, 7ell* | | **`LED_OFF`** | Turns the LED OFF | *turn off led, sakker, taffi* | | **`STATUS`** | Returns system state | *status, state, chnowa status* | ## 🧠 Machine Learning Pipeline * **Model Architecture:** Small Dense Neural Network * **Data Input:** Bag-of-Words (BoW) vectorization * **Model Output:** 3-class Softmax activation * **Framework:** TensorFlow ➔ TensorFlow Lite (`.tflite`) ➔ STM32Cube.AI (X-CUBE-AI) * **Deployment:** STM32Cube.AI compiles the `.tflite` model into an optimized C library integrated directly i …