📦 mattmireles/gemma-tuner-multimodal

Train Multimodal AI on Your Mac!

Want to fine-tune powerful AI models without needing an expensive cloud server? This toolkit lets you train Gemma models on your own Apple Silicon Mac, turning your laptop into a high-performance machine. Whether you are working with text, images, or audio, you can now run complex fine-tuning jobs locally using PyTorch and Metal Performance Shaders. It solves the hurdle of needing massive hardware by allowing efficient, native training on Apple chips, perfect for custom vision or audio projects. Take full control of your AI development pipeline and keep your data private by building everything directly on your own device.

📰 https://news.ycombinator.com/item?id=47680309

🆔 @hackernewsgithubprojects

📦 santifer/career-ops

Automate Your Job Search with AI

Are you tired of manually tracking job applications and tailoring resumes for hours on end? Meet an open-source AI-powered command center that transforms your job search into a streamlined pipeline. By using this system, you can automatically evaluate job offers using a structured scoring system and generate perfectly tailored, ATS-optimized resumes for every role. It features a custom dashboard for tracking progress and leverages smart agentic workflows to scan portals and perform deep research. Stop the repetitive grind and let intelligent automation handle the heavy lifting while you focus on landing your next big career move today.

📰 https://news.ycombinator.com/item?id=47697285

🆔 @hackernewsgithubprojects

📦 plasma-umass/coz

Stop Guessing, Start Optimizing: Introducing Coz

Are you tired of spending hours optimizing code that does not actually improve your application speed? Meet Coz, a revolutionary causal profiler that changes everything by predicting exactly how much your performance will improve before you even change a single line of code. Unlike traditional profilers, Coz conducts virtual performance experiments to show you the direct impact of your efforts on throughput and latency. By using progress points to track work, it reveals the true bang for your buck, helping you focus only on the high-impact code that truly counts. Start coding smarter and stop wasting time on blind optimizations.

📰 https://news.ycombinator.com/item?id=47697295

🆔 @hackernewsgithubprojects

📦 farzaa/clicky

Meet Clicky: Your New AI Screen Companion

Ever wondered what it would be like to have a personal AI tutor living right on your screen? Meet Clicky, an open-source Mac companion that acts like a real-time teacher sitting next to your cursor. It uses advanced screen recording to see exactly what you are working on while listening to your voice through push-to-talk. By leveraging Claude, AssemblyAI, and ElevenLabs, it can actually point to specific interface elements and talk you through complex tasks. It is a brilliant way to supercharge your workflow and get instant, interactive help whenever you need it most. Give it a try.

📰 https://news.ycombinator.com/item?id=47699838

🆔 @hackernewsgithubprojects

📦 keychron/keychron-keyboards-hardware-design

Keychron Reveals Their Keyboard Engineering Secrets

Have you ever wondered what goes into designing your favorite mechanical keyboard? Keychron just opened up their digital vault, sharing official hardware design files for over a hundred of their most popular models. This treasure trove provides detailed CAD files like STEP and DXF formats for everything from internal stabilizers to intricate outer cases. Whether you are a custom builder modding your deck or a curious tinkerer studying keyboard architecture, these blueprints offer an unprecedented look at industrial engineering. It is an incredible resource for understanding precision hardware manufacturing, so dive in, explore these professional designs, and level up your custom projects.

📰 https://news.ycombinator.com/item?id=47699911

🆔 @hackernewsgithubprojects

📦 pcruz1905/hls-restream-proxy

Fix Your IPTV Streams in Jellyfin or Plex Easily

Struggling to get live streams working in your media server? This lightweight proxy is a game-changer for anyone using Jellyfin, Emby, or Plex. Many streaming sources fail because they require specific headers like User-Agent or Referer that media servers simply do not send. This tool acts as a clever middleman, injecting the necessary headers and rewriting playlists so your streams stay active. It even includes a handy detector tool to identify exactly which headers your specific streams need. Stop fighting with expired tokens and broken links, and start enjoying a seamless and automated live television experience today.

📰 https://news.ycombinator.com/item?id=47689165

🆔 @hackernewsgithubprojects

📦 wong2/diffx

Supercharge Your Coding Agent Workflow with Diffx

Stop wasting time manually auditing every change made by your automated coding agents. Meet your new best friend for local code reviews. This powerful tool acts as a bridge, giving you immediate visibility into what your AI assistant is actually modifying within your local files. By generating clear, readable diffs directly in your terminal, it allows you to quickly spot errors or unexpected refactors before they ever hit your main codebase. It makes the entire AI development cycle safer and more transparent, ensuring you stay in full control. Give it a try to experience a smarter, smoother workflow today.

📰 https://news.ycombinator.com/item?id=47638679

🆔 @hackernewsgithubprojects

📦 bitterbot-ai/bitterbot-desktop

Bitterbot: The AI Agent That Dreams and Earns

Most AI agents are just temporary wrappers, but Bitterbot is a persistent, local-first companion designed to grow alongside you. By integrating a biological memory system and a unique dream engine, this agent processes experiences, consolidates knowledge, and evolves its personality while you sleep. Beyond just chatting, it masters complex workflows and publishes them to a decentralized peer-to-peer marketplace, allowing your agent to earn revenue autonomously. It even manages its own wallet for micropayments and resource access. Whether it is browsing the web or coding, Bitterbot builds a truly living identity. Start your own agent node today and join the evolving network.

📰 https://news.ycombinator.com/item?id=47654821

🆔 @hackernewsgithubprojects

📦 kimwalisch/primesieve

The Fastest Way to Generate Prime Numbers

Ever wonder how to generate massive amounts of prime numbers in seconds? Meet Primesieve, a high-performance, open-source tool built for incredible speed. It uses the highly efficient segmented sieve of Eratosthenes algorithm, carefully optimized to match your specific CPU cache sizes for maximum throughput. Whether you need to count primes, find specific values, or perform system stress tests, this library handles it all with ease across multiple CPU cores. It is a fantastic choice for developers and math enthusiasts alike who need raw computational power. Dive into this project today and supercharge your next math-heavy application effortlessly.

📰 https://news.ycombinator.com/item?id=47701677

🆔 @hackernewsgithubprojects

📦 nickvido/us-code

Track Changes in US Federal Law Using Git

Ever wondered exactly how federal law changes over time? The US Code repository turns the entire United States Code into a searchable Git project. By converting statutory text into structured Markdown, it lets you use familiar developer tools to analyze legislation. You can use standard commands like git diff to see precise changes between different years or congresses, and git blame to track when specific legal provisions were modified. This project makes complex legislative history transparent and easy to explore. It is a powerful way to visualize the evolution of American law, providing a clear snapshot of history.

📰 https://news.ycombinator.com/item?id=47625366

🆔 @hackernewsgithubprojects

📦 asciimoo/hister

Build Your Own Private Search Engine

Ever wanted to take total control over your web searching experience without relying on big tech tracking your every move. Meet Hister, a powerful tool that lets you host your very own private search engine. Instead of handing your data to someone else, this lightweight software lets you aggregate results from multiple sources locally. By deploying it on your own server, you gain a custom, privacy focused gateway to the internet that runs exactly how you want. It is a fantastic way to reclaim your digital sovereignty while keeping your queries private. Give it a try and own your search.

📰 https://news.ycombinator.com/item?id=47702208

🆔 @hackernewsgithubprojects

📦 0din-ai/ai-scanner

Is Your AI Truly Safe?

Have you ever wondered if the artificial intelligence you are using is actually safe from manipulation? This project introduces a powerful security tool designed to stress test AI models for vulnerabilities. By utilizing the robust NVIDIA garak framework, it systematically probes your language models to identify potential risks like prompt injection or harmful content generation. This is essential for developers who need to harden their systems before deployment. Instead of guessing, you get a clear look at how your model handles attacks, ensuring a safer experience for everyone. Start scanning your models today to keep your AI infrastructure secure.

📰 https://news.ycombinator.com/item?id=47702784

🆔 @hackernewsgithubprojects

📦 jfalcou/eve

Supercharge Your C++ Performance with EVE

Want to make your C++ code fly? Meet the Expressive Vector Engine, a powerful library designed to simplify SIMD programming. SIMD allows your processor to handle multiple data points at once, but writing it manually is often a nightmare. This repository solves that by providing a high-level, type-based wrapper that works seamlessly across various architectures like Intel and ARM. By leveraging modern C++ twenty features, it brings efficient, low-level performance to your projects without the usual headaches. It is the perfect playground for developers looking to push the envelope of speed and clean, expressive software design today.

📰 https://news.ycombinator.com/item?id=47703090

🆔 @hackernewsgithubprojects

📦 vincent-uden/miro

Miro: The Fast, Keyboard-Driven PDF Viewer You Need

Are you tired of clunky PDF readers that slow you down? Meet Miro, a high-performance native PDF viewer built for Windows and Linux. Whether you prefer Vim-like keybindings or custom shortcuts, this tool gives you total control over your reading experience. It handles multiple documents in tabs, features a clean dark mode, and even includes automatic hot-reloading for your LaTeX or Typst projects, making it perfect for power users and developers alike. If you need a fast, keyboard-driven way to manage your documents, give Miro a try and streamline your entire reading and document workflow starting today.

📰 https://news.ycombinator.com/item?id=47703079

🆔 @hackernewsgithubprojects

📦 simdjson/simdjson

Parsing JSON at Gigabytes Per Second!

Are you tired of slow JSON parsing dragging down your high-performance applications? Meet the powerhouse library that changes everything by parsing gigabytes of data every single second. By leveraging advanced SIMD vector instructions and microparallel algorithms, this project achieves speeds significantly faster than traditional parsers while ensuring strict validation of every single byte. It automatically selects the best CPU-optimized implementation at runtime, making it incredibly efficient for demanding workloads like databases and runtimes. If you need lightning-fast, production-grade JSON processing, this library is your ultimate solution to unlock massive speed gains in your next big project.

📰 https://news.ycombinator.com/item?id=47703684

🆔 @hackernewsgithubprojects

📦 hkuds/deeptutor

DeepTutor: Your Personal Agent-Native AI Learning Assistant

Struggling to manage your studies and research in one place? Meet DeepTutor, an agent-native assistant designed to revolutionize how you learn. Instead of standard chatbots, it offers persistent tutor bots that remember your progress and evolve with you. Whether you need to generate quizzes, conduct deep research, or create complex math animations with its integrated tools, DeepTutor keeps your entire workflow in one unified space. With a powerful command line interface and support for custom knowledge bases, it turns information into actionable mastery. Stop juggling scattered notes and let this intelligent system guide your personal learning journey today.

📰 https://news.ycombinator.com/item?id=47703911

🆔 @hackernewsgithubprojects

📦 lamm-mit/scienceclaw

ScienceClaw: Autonomous AI Agents for Scientific Discovery

Ever wondered if AI could handle the heavy lifting of scientific research on its own? Meet ScienceClaw, an open-source framework that deploys autonomous agents to conduct complex investigations without central coordination. By utilizing over three hundred interoperable scientific tools, these agents chain workflows to generate testable hypotheses and produce immutable research artifacts. The system uses a clever reactor mechanism to broadcast unmet data needs, allowing peer agents to automatically fill in knowledge gaps. It is a powerful way to accelerate discovery by building a traceable, collaborative ecosystem where artificial agents continuously expand the boundaries of human scientific inquiry.

📰 https://news.ycombinator.com/item?id=47704025

🆔 @hackernewsgithubprojects

📦 tronghieuit/tiny-tts

The World's Smallest AI Voice Model

Have you ever wanted to run high-quality text to speech directly on your device without needing a massive server or a powerful graphics card? Meet the smallest English speech synthesis model ever built, featuring a tiny footprint of just over one million parameters. By utilizing an optimized ONNX format, this incredible engine delivers crisp audio while remaining lightweight enough to run effortlessly on standard CPUs and embedded systems. Whether you are building real-time applications or working with limited hardware, this project proves that extreme efficiency does not have to sacrifice performance. Grab this tool to start creating your own lightweight voice applications today.

📰 https://news.ycombinator.com/item?id=47689557

🆔 @hackernewsgithubprojects

📦 spring-ai-community/spring-ai-playground

Build and Test AI Tools Instantly with Spring AI Playground

Are you tired of constantly restarting your applications just to test a small change in your AI tools? Meet the Spring AI Playground, a powerful self-hosted environment designed to supercharge your development workflow. This platform lets you build, test, and expose reusable tools using a low-code approach that instantly bridges the gap between static chats and real-world actions. With its built-in server, you can create and refine tools in real-time, then seamlessly integrate them with AI agents without redeployment. It is the perfect hub for experimenting with agentic chat and RAG workflows. Start building smarter, more responsive AI systems today.

📰 https://news.ycombinator.com/item?id=47703920

🆔 @hackernewsgithubprojects

📦 milanm/autograd-engine

Build a GPT from Scratch in Pure C#

Ever wondered how ChatGPT actually learns to generate text? This repository is a brilliant educational project that implements a complete GPT language model using only pure C# with absolutely zero external dependencies. In just six hundred lines of code, it replicates the exact algorithms powering massive AI models, including the transformer architecture and backpropagation. You get to see the magic of a custom autograd engine that tracks operations for learning. It is a fantastic, hands-on way to master neural networks, tokenization, and attention mechanisms from the ground up. Go explore this code to unlock the secrets behind modern artificial intelligence.

📰 https://news.ycombinator.com/item?id=47704687

🆔 @hackernewsgithubprojects

📦 luce-org/luce-megakernel

Stop Waste: How This Megakernel Crushes Power Efficiency

Did you know that the efficiency gap between NVIDIA hardware and Apple Silicon is often a software issue, not a hardware one? This repository showcases a pioneering megakernel that fuses all twenty-four layers of a hybrid model into a single CUDA dispatch. By eliminating the constant overhead of hundreds of separate kernel launches per token, it drastically reduces unnecessary synchronization and weight re-fetching. On a five-year-old graphics card, this specialized approach achieves double the throughput and matches the efficiency of modern chips, proving that smarter code can unlock hidden potential in the hardware you already own today.

📰 https://news.ycombinator.com/item?id=47691182

🆔 @hackernewsgithubprojects