kavishka-dot

kavishka-dot / libminerva

Public

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

arduino atmega avr c embedded-ml
13
2
100% credibility
Found May 12, 2026 at 13 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
C
AI Summary

MINERVA is a compact C library that enables running encrypted, integrity-protected machine learning models on low-resource microcontrollers with defenses against tampering, side-channel leaks, and fault injection.

How It Works

1
📰 Discover Secure Tiny AI

You hear about Minerva, a way to run smart predictions safely on tiny gadgets like Arduino without needing big computers.

2
📥 Get the Free Toolkit

Download the simple files to your computer to start building secure smart helpers for your projects.

3
🔧 Pick Your Gadget and Brain Size

Choose the type of small device you're using and how big a smart model you want it to handle, like for sensor readings.

4
🧠 Prepare a Sample Smart Model

Use the ready example brain that detects patterns, or create your own simple one for what you need.

5
🔒 Secure It with a Private Password

Create a special secret code once to lock your smart model so only your gadget can unlock and trust it.

6
⚙️ Add to Your Gadget Project

Mix the protected brain into your gadget's code, set it up, and connect your sensors.

7
🚀 Run and Test Predictions

Power on your gadget, feed it real-world data, and watch it make safe, reliable decisions.

✅ Trusted Smart Results

Your tiny gadget now delivers secure, tamper-proof insights every time, keeping everything certain and private.

Sign up to see the full architecture

6 more

Sign Up Free

Star Growth

See how this repo grew from 13 to 13 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is libminerva?

Libminerva is a minimal C inference engine for running encrypted, integrity-verified neural networks on MCUs down to the ATmega328P. It lets you deploy robust ML models—like MLPs, 1D CNNs, or binary nets—with military-grade security: ChaCha20 encryption, BLAKE2s authentication, constant-time execution, and anti-glitch protection. A Python compiler turns your trained .npz model into C arrays; integrate with init/run calls for ~20Hz inference using ~14KB flash and 960B RAM.

Why is it gaining traction?

Unlike bloated frameworks, libminerva enforces zero dynamic allocation, side-channel resistance via blinded LUTs and output session MACs, and tamper detection that halts on errors. Developers grab it for fitting secure, verifiable AI on AVR/STM32 without leaks—think sensor classification that survives physical attacks. Amid github minerva searches (airbnb archives, myrient workers, math libs), this stands out for MCU-specific hardening.

Who should use this?

Embedded engineers building secure edge AI on ATmega328P or STM32 for IoT anomaly detection, fault classification, or authenticated predictions. Ideal for devs shipping battery-powered sensors where model tampering or side-channel exploits could brick devices. Skip if you need GPU-scale training—focus is tiny, verified inference.

Verdict

Promising for secure MCU ML, with solid docs, 26 passing tests, and ready examples, but 13 stars and 1.0% credibility score signal early maturity—test thoroughly before production. Worth a prototype if integrity-verified inference is non-negotiable.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.