kavishka-dot

kavishka-dot / filum

Public

Pure-C federated learning library for MCU-class edge devices over LoRa. STM32 + SX1276.

c99 edge-ai embedded federated-learning iot
12
0
100% credibility
Found May 10, 2026 at 12 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
C
AI Summary

Filum lets tiny low-power devices train shared AI models collaboratively over radio without exposing private data, coordinated by a host computer.

How It Works

1
📖 Discover Filum

You find Filum, a clever way for tiny battery-powered gadgets to team up and get smarter together using radio signals, keeping their private sensor info safe.

2
💻 Try the demo

On your regular computer, you run a simple test that shows gadgets learning from made-up data, proving it works without any real hardware.

3
🔧 Prepare your gadgets

You load the learning brains onto your small sensor devices and set up a coordinator on a Raspberry Pi or computer to guide them.

4
📡 Link them by radio

Your devices start chatting with the coordinator over long-range radio waves, sharing only safe learning updates.

5
🚀 Launch learning rounds

The coordinator signals everyone to train on their local data, and devices send back improvements to make the whole group smarter.

🎉 Smarter devices network

Your fleet of gadgets now collaborates to build a shared smart model, getting better at tasks like sensing patterns while staying private and power-efficient.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 12 to 12 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 filum?

Filum is a pure-C federated learning library for MCU-class edge devices over LoRa, letting STM32 shards train models on local sensor data and share sparse gradient updates with a Linux Herald coordinator. It handles aggregation, model deltas, and rounds without sending raw data, fitting in 128KB RAM with no heap or Python. Search filum filderstadt or filum terminale? This filum delivers edge federated learning for devices.

Why is it gaining traction?

Zero dynamic allocation and LoRa-optimized payloads mean 5.5x compression via Q8 sparse gradients, fitting SF7 packets perfectly. Built-in differential privacy, ECDH encryption, and presets from 2KB (STM32F103) to 1MB make it dead simple for constrained setups. End-to-end demos run in-memory, no hardware needed, hooking devs tired of bloated ML frameworks.

Who should use this?

Embedded devs building battery-powered sensor fleets for smart agriculture or environmental monitoring over LoRa. Suited for teams wanting privacy-focused federated learning on STM32 edges, like filum krankheit tracking in remote clinics or filumi kinderzentrum prototypes. Avoid if you need conv nets or GPU scale.

Verdict

Solid start for pure-C federated learning library on edge devices—great docs, 9 tests, CI coverage—but 12 stars and 1.0% credibility scream early alpha. Prototype with it now; production needs field hardening.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.