General Description
EmbedIA is a lightweight machine learning framework for deploying neural networks and classical ML models on microcontrollers with constrained resources. It converts trained models into efficient C/C++ code to run inference directly on embedded hardware, making edge AI practical and portable.
This project was part of my supervised professional practice in which I worked on the export pipeline from a quantized neural network into an EmbedIA FULL_QUANT8 project. The goal was to take a model developed in Python, optimize it for embedded execution, and generate a compact C/C++ deployment that could run on MCU-class hardware without losing the expected behavior.
Personal Contribution
My contribution focused on preparing the quantized model for embedded export, validating the generated runtime, and fixing the compatibility issues that surfaced during conversion. I worked on the export flow, corrected the Dense/template logic when precision dropped, and verified that the generated project compiled and executed correctly on the provided example dataset.
This was a very valuable experience because it showed me that embedded AI is not only about model accuracy in Python, but also about quantization, memory constraints, runtime compatibility, and deployment reliability. The process taught me how important it is to optimize both the model and the generated C implementation so the system can run efficiently on real hardware.
Gallery
EmbedIA project image