📦 ghentlesteve/dhis2-immunisation-audit

Why Smart Models Lost to a Simple Guess

A basic rule of just repeating last year’s numbers beats complex machine learning models when analyzing this public health dataset. This project pulls six years of vaccine records for thirteen districts, then acts like a detective to find hidden errors. It discovered that most data points were exact copies of the previous year, meaning the information wasn’t real. It also found that entire months had been swapped in order. Because the data was so repetitive, a simple forecast outperformed advanced statistical tools. The real lesson here is that checking your data for weird patterns matters more than choosing the fanciest algorithm.

🆔 @hackernewsgithubprojects

📦 hongnoul/hwatu

hwatu: The 35ms Browser for AI Agents

Stop guessing if your code actually works. hwatu lets AI coding agents verify their changes in thirty-five milliseconds, replacing vague visual checks with hard, measurable pixel-perfect scores. It’s a warm, invisible browser that never steals your focus, handing back exact data instead of screenshots. The best part? If it hits a wall, you can instantly pop the live session into a real window to take over. It’s the end of the vibes check, and the start of verified code.

📰 https://news.ycombinator.com/item?id=49097727

🆔 @hackernewsgithubprojects

📦 0sec-labs/0

Software That Hunts Its Own Bugs

Software now hunts its own bugs before they ever reach a hacker. Zero is an open-source security harness from Swiss Applied AI and Cybersecurity Research Lab that uses artificial intelligence to autonomously find and fix vulnerabilities in complex systems. It acts like a tireless digital security guard, constantly scanning codebases and testing weaknesses around the clock without human fatigue. This is useful because it shifts the burden of security from manual, error-prone audits to automated, continuous defense. The project is interesting because it aims to make software secure itself, allowing developers to focus on creation while the system handles the heavy lifting of threat detection and repair.

📰 https://news.ycombinator.com/item?id=49891508

🆔 @hackernewsgithubprojects

📦 chrisryugj/kordoc

kordoc: The Korean Document Translator

kordoc can read a Korean government file and turn it into clean text without losing a single table cell. Built by a former civil servant, it solves a unique problem where standard converters destroy the complex layouts of official forms. It reads the older Hancom formats, modern office files, and even scanned papers, turning them into simple text that computers can actually use. This means you can finally compare old and new drafts or fill out forms automatically, all without needing the original software. It is the rare tool that respects the messy reality of bureaucratic paperwork.

🆔 @hackernewsgithubprojects

📦 zoemaestra/hair-electrolysis-machine-usb-c

DIY Hair Removal Machine Under Twenty Dollars

Build a portable hair removal device that runs on a small lithium polymer battery and a tiny microcontroller. This open hardware project redesigns a classic electrolysis circuit onto a single sided board, cutting the cost to about twenty dollars and removing the need for soldering entirely. You simply plug in a display and a battery to get a working unit with a digital interface and safety checks. It turns a complex medical grade tool into an affordable DIY build, letting you assemble your own hair removal system from parts ordered online.

🆔 @hackernewsgithubprojects

📦 whiskerrs/whisker

Whisker: Native Apps in Rust

Whisker is the app framework that lets you write one codebase in Rust to ship truly native apps on iOS, Android, web, and desktop. The standout feature here is the hot reload system, which patches your running app in under a second while keeping your state intact. This means you can tweak a button’s color or logic, save the file, and see the change instantly without restarting the app or losing your data. It uses a fine-grained reactive runtime to update only what changed, avoiding the overhead of a virtual DOM.

📰 https://news.ycombinator.com/item?id=49892519

🆔 @hackernewsgithubprojects

📦 eceo-epfl/deepreefmap

DeepReefMap: 3D Mapping from a GoPro

DeepReefMap turns a standard handheld underwater video into a detailed three-dimensional map of a coral reef. By analyzing frames from a GoPro, the software reconstructs the physical shape of the environment and simultaneously labels every point with specific coral species. This allows researchers to measure benthic cover and study ecosystem health without expensive equipment. The tool processes the footage to generate a point cloud and an ortho mosaic, making complex marine data accessible. It is a powerful open-source solution for documenting reef structures from simple video footage.

🆔 @hackernewsgithubprojects

📦 rubys/roundhouse

Roundhouse: One Ruby App, Ten Different Languages

One Ruby codebase can now compile into ten completely different languages without changing a single line of source code. Roundhouse reads your Rails application and treats the deployment target as a simple build flag, producing standalone Rust, Go, or Swift binaries from the exact same Ruby you already have. It works by inferring types from Rails conventions rather than requiring annotations, meaning it analyzes your code without booting the app or touching a database. This lets you keep your Ruby codebase while deploying to native targets that run significantly faster than standard Ruby.

🆔 @hackernewsgithubprojects

📦 bingeljell/image-to-3dlab

Image to 3D Lab: Local 3D Generation

Image to 3D Lab turns a single flat photograph into a fully textured, three-dimensional model using only the hardware already inside your computer. This local-first tool runs entirely on Apple Silicon or NVIDIA graphics cards, meaning no data leaves your machine and no cloud services are involved. The standout feature is its provenance tracking, which creates a digital receipt for every model, detailing exactly which AI components were used and what license restrictions apply. This is crucial for developers who need to ensure their game assets are legally safe for commercial use. By handling the complex math and licensing locally, it gives creators full control and transparency.

🆔 @hackernewsgithubprojects

📦 ingelibre/ingetrazo

IngeTrazo: The Linux SketchUp

Trace 3D buildings right on your Linux desktop with IngeTrazo, a free modeler that finally gives architects a native tool for their system. It works just like SketchUp, letting you push and pull shapes to build complex structures, but it adds a unique layer of engineering intelligence. You can tag your digital walls and slabs so the software calculates exact material quantities for real budgets. This bridges the gap between simple sketching and professional construction planning, all without leaving your favorite operating system. It is the missing piece for engineers who refuse to switch platforms just to do their job.

🆔 @hackernewsgithubprojects

📦 basecamp/house-skills

House Skills: The Playbook for AI Developers

House Skills is the shared playbook that turns generic AI coding assistants into specialized, reliable team members. It solves the problem of vague instructions by providing specific, tested modules for tackling complex tasks like automating security fixes or organizing chaotic project updates. The most surprising part is how it handles trust. Instead of blindly following prompts, it teaches agents to treat external data as untrusted and to verify every command before executing it. This ensures your AI helper doesn't accidentally break production systems or leak sensitive data. It is a clear guide for developers who want to use AI safely and effectively.

🆔 @hackernewsgithubprojects

📦 codingnoye/gksdud

gksdud: The Snappy Mac Switcher

Switch your Mac input source without the annoying delay. gksdud is a tiny utility that fixes the sluggish switching between Korean and English. It works by remapping your key to a hidden system trigger, so the change happens the instant you press it. No complex tools needed. It even remembers your capital letters, which is a lifesaver. It replaces the standard menu bar icon to save space. If you type in multiple languages, this makes it feel instant. It is a clever, low-friction fix for a daily annoyance. Give it a try. It is surprisingly good.

🆔 @hackernewsgithubprojects

📦 wuji-technology/wuji-mjlab

wuji-mjlab

Teach a robotic hand to perform incredibly complex in hand manipulations like spinning a pen or rotating a cube with high precision. This project provides a complete reinforcement learning and simulation to real deployment pipeline designed specifically for the multi fingered Wuji Hand. By training artificial intelligence policies using proximal policy optimization inside a physical simulator, it bridges the gap between digital training and physical execution. The system transfers trained neural network policies directly onto real robotic hardware for closed loop control. It is an amazing resource for robotics researchers looking to master complex, real world dexterous manipulation.

🆔 @hackernewsgithubprojects

📦 scarletkc/seiso

Seiso: The Linter for AI Docs

Seiso is the linter that keeps AI-written project docs from lying to your team. It treats documentation like code, enforcing a single convention so every fact has one clear home and no page holds stale values. Since coding agents read these files as context for the next change, a single outdated version number can mislead both humans and bots. Seiso spots those inconsistencies before they cause real problems, giving you a clean, reliable source of truth. It’s the simple way to make sure your project’s written rules actually match what’s happening in the repo. Stop guessing if your docs are current.

🆔 @hackernewsgithubprojects

📦 bendlang/bend

Bend: The Language That Stops AI Bugs

Stop trusting your code by sight and start proving it works. This is how the project called Bend stops artificial intelligence from breaking your software. You write specific rules, called laws, like a player can never win. The compiler then mathematically verifies your code against those laws. If the AI tries to add a feature that breaks the rule, the compiler rejects it instantly. It forces the machine to fix its own mistakes before you ever see them. It is a fast language that blocks bugs through proof, making trust automatic. If you want your AI to build things you can actually rely on, this is the tool to watch.

📰 https://news.ycombinator.com/item?id=49746203

🆔 @hackernewsgithubprojects

📦 shoyrulove-dev/purehub

26 Tools, One App

PureHub bundles twenty-six distinct utility tools into a single offline-first app, eliminating the need to install and manage multiple separate programs. Instead of scattering your data across various services, this open-source project processes everything locally on your device, ensuring your information never leaves your hands. The most impressive feature is its document scanner, which extracts text directly from your camera using on-device intelligence, turning physical pages into editable digital files without sending a single pixel to the cloud. It replaces a cluttered app drawer with a calm, ad-free workspace for measuring, scanning, and organizing your day.

🆔 @hackernewsgithubprojects

📦 typesafe-ai/typesafe-sdk-js

TypeSafe AI JavaScript SDK: Get typed answers from text

Turn raw customer messages into structured, typed data by defining your own questions right in your code. The TypeSafe AI JavaScript SDK lets you send text like a billing complaint to a server, which then reads it and answers specific questions you set up, such as whether it is about money or a technical glitch. The coolest part is that your code automatically knows the shape of the answer, so you stop guessing what the server returns and start working with clean, predictable data. It handles the tricky networking stuff for you, like retrying failed calls and managing your API key safely.

🆔 @hackernewsgithubprojects

📦 typesafe-ai/typesafe-sdk-python

How TypeSafe AI turns messy text into clean data

TypeSafe AI is the Python library that turns messy real-world text into structured data your code can actually use. Instead of relying on fragile patterns that break when a word changes, you hand it a document and a simple question, like asking to categorize a customer complaint, and it returns a clear, typed answer. The SDK handles the connection, retries, and validation in the background, so you get reliable results without writing complex parsing logic. It is built for developers who need consistent output from unpredictable language.

🆔 @hackernewsgithubprojects

📦 moonshotai/kimi-k2.5

Kimi K2.5: The AI That Sees and Codes

Kimi K2.5 is the open-source multimodal model that finally lets artificial intelligence see, reason, and write code all at once. It’s built on a massive mixture of experts architecture, meaning it has a trillion parameters but only activates thirty-two billion when you need answers, which keeps it efficient without sacrificing brainpower. The real headliner, though, is its ability to look at a UI design or a video workflow and then write the actual code to build it, not just describe it. It’s a native visual agent that bridges the gap between what you see and what gets done.

🆔 @hackernewsgithubprojects

📦 agentara/skills

Skills: The AI Toolbox You Didn't Know You Needed

Turn your coding assistant into a creative powerhouse with Skills, a personal library of installable extensions that give Claude Code entirely new capabilities. Instead of just writing code, these prompt packs teach your AI how to design cinematic video storyboards, interpret medical lab reports, or even predict World Cup outcomes. The coolest part is the entertainment category, where a single skill spawns parallel subagents to analyze live news and market data for real-time sports forecasting. It’s a clever way to turn a standard terminal tool into a specialized creative and analytical partner.

🆔 @hackernewsgithubprojects