{"slug": "i-ran-mnist-on-an-esp32-c3-without-tensorflow-tflite-or-any-ml-runtime", "title": "I ran MNIST on an ESP32-C3 without TensorFlow, TFLite, or any ML runtime", "summary": "The author successfully ran MNIST digit recognition on an ESP32-C3 microcontroller by compiling a neural network directly into a C header file and executing it as firmware, without using TensorFlow, TFLite, or any ML runtime. The project evolved from a simple perceptron experiment into a cross-platform CLI tool that generates minimal, deterministic inference pipelines for constrained edge devices. The author emphasizes that the toolchain prioritizes stability and portability over competing with full ML frameworks, focusing on how small a usable neural inference stack can become while remaining practical.", "body_md": "I ran MNIST digit recognition on an ESP32-C3 — without TensorFlow, TFLite, or any ML runtime.\nThe neural network is compiled directly into a C header and executed as firmware.\nIt started, like many questionable engineering projects, with curiosity about a perceptron.\nI wasn’t trying to build anything serious. Just a small experiment to understand how far I could push a minimal neural network implementation without relying on any framework.\nThat experiment didn’t stay small for long.\nOnce the basic forward/backward pass worked, I started wondering whether the same approach could be useful outside of toy problems. Specifically, I had recently built a ratiometric thermometer system, and it quickly became clear that the analog measurements were noisy and sensitive to supply variation.\nThat’s where the idea clicked: instead of fitting everything with fixed equations, what if I used a lightweight neural network as an inference layer for correction?\nNot training in the field — just inference on constrained hardware.\nThat idea turned into a CLI tool, not a library. Not a framework. A command-line tool.\nThe goal was simple:\nNo heavy dependencies. No runtime frameworks required on the target side, just portable inference.\nOver time, this evolved into a cross-platform tool supporting Linux, macOS, and Windows, with a fully deterministic inference pipeline.\nAt some point I needed a sanity check, so I ran MNIST through it.\n~59,999 training samples~1,000 test samples\nNothing exotic — just a standard benchmark to validate the pipeline end-to-end.\nThe interesting part wasn’t accuracy. It was stability.\nTraining, export, and inference all behaved consistently across platforms. No drift between environments, no dependency on external runtimes, and no hidden assumptions about floating-point behavior beyond what was explicitly defined.\nFor a homegrown stack, that was the real milestone.\nThe original motivation was never image classification, it was edge inference under constraints:\nIn that context, most modern ML stacks are overkill. Even lightweight frameworks often assume too much infrastructure.\nSo the focus shifted toward something more specific:\nA minimal neural network runtime that could be compiled down and deployed without dependencies.\nToday, the project is essentially three things:\nA minimal neural network training engine\nA model exporter that produces standalone inference graphs\nA CLI runtime for cross-platform execution\nIt is not meant to compete with full frameworks.\nIt is meant to answer a narrower question:\nHow small can a usable neural inference stack get while remaining practical?\nhttps://github.com/AlexRosito67/xyron-mnist-esp32\nThis started as an experiment to understand perceptrons and a thermometer, it ended as a small toolchain for compiling neural networks into embedded firmware.\nAnd I'm still not entirely sure where the boundary between \"software\" and \"firmware\" is anymore.", "url": "https://wpnews.pro/news/i-ran-mnist-on-an-esp32-c3-without-tensorflow-tflite-or-any-ml-runtime", "canonical_source": "https://dev.to/alexrosito67/i-ran-mnist-on-an-esp32-c3-without-tensorflow-tflite-or-any-ml-runtime-1cjk", "published_at": "2026-05-20 06:07:25+00:00", "updated_at": "2026-05-20 06:35:39.170855+00:00", "lang": "en", "topics": ["machine-learning", "hardware", "developer-tools", "open-source", "research"], "entities": ["ESP32-C3", "MNIST", "TensorFlow", "TFLite"], "alternates": {"html": "https://wpnews.pro/news/i-ran-mnist-on-an-esp32-c3-without-tensorflow-tflite-or-any-ml-runtime", "markdown": "https://wpnews.pro/news/i-ran-mnist-on-an-esp32-c3-without-tensorflow-tflite-or-any-ml-runtime.md", "text": "https://wpnews.pro/news/i-ran-mnist-on-an-esp32-c3-without-tensorflow-tflite-or-any-ml-runtime.txt", "jsonld": "https://wpnews.pro/news/i-ran-mnist-on-an-esp32-c3-without-tensorflow-tflite-or-any-ml-runtime.jsonld"}}