Microduck is a robot about 25 cm tall and 800 g. This repo is its brain: a set of Rust daemons that run the control loop, drive fifteen servos, handle the radios and camera, and install software updates.

The walking and other motions come from neural policies. They are trained in a separate repo, microduck_rl, using MuJoCo and PPO. They are exported to ONNX, and this repo loads them and runs them on the robot.

The system is split into small programs that talk over one JSON-RPC contract on Unix sockets. The app, the console, the gamepad and your own scripts all send the same calls. Updates are signed, checked for health, and can be rolled back. There is also a simulated duck in MuJoCo, so you can try the real daemons without hardware.

Read more on interesting repos · GitHub

World Monitor puts many streams of world news and data on one screen. Instead of checking separate news sites, market tickers and tracking tools, you get feeds, maps and panels together. An AI layer condenses the news into short briefs.

The front end is vanilla TypeScript with Vite. It has two map views: a 3D globe built on globe.gl and a flat WebGL map built on deck.gl. The AI parts can call Ollama, Groq or OpenRouter, and some models run in the browser through Transformers.js. The README says everything can run locally on Ollama with no API keys.

One codebase produces several variants: world, tech, finance, commodity, happy and energy. It also ships a Tauri desktop app. For programs and agents, it offers an MCP server, a REST API, a CLI and client libraries for Python, Ruby and Go. The README does not name direct alternatives.

Read more on interesting repos · GitHub

Every WiFi router fills a room with radio waves, and people disturb those waves when they move or breathe. RuView captures these disturbances as Channel State Information (CSI) from ESP32 boards that cost about $9 each. It turns them into data about who is in a room and what they are doing.

Signal processing handles the basics: band-pass filters on the signal phase give breathing and heart rate, and phase changes flag motion and falls. A small pretrained model from Hugging Face handles presence and embeddings, and a fallback based on phase variance works with no model. It runs on edge hardware with no cloud, and learns each room locally.

Compared with cameras or wearables, nothing is filmed and nobody wears anything. The README is open about limits: the on-device pose model is a first cut with low accuracy, the unified RF world model is still validated only on synthetic data, and an older 100% presence figure has been retracted. The repo also ships an npm toolkit, MetaHarness, so AI agents can help with setup, calibration and verification.

Read more on interesting repos · GitHub

Beat Claude is a public hiring challenge run by the marketing agency Single Grain. Candidates pick a task for a role, build a real work artifact, and submit it with evidence. The aim is to find people who add judgment and proof to what an AI tool produces from the brief alone.

The repo holds the candidate briefs, the public scoring guide, fixture datasets and a validator script. Reviewers score answers against private benchmarks that are not published, to discourage rubric gaming. Strong submissions may get a follow-up exercise or a live walkthrough.

It is not software you run for its own sake. It is a hiring process in the open, and its design is worth reading if you want to see how a company tests AI-assisted work. Tools of any kind are allowed, but you must disclose how you used them.

Read more on interesting repos · GitHub