Show HN: FizzBuzzAI – The Most Inefficient FizzBuzz Solution Ever Made https://bit.ly/3FdB3yZ

Show HN: FizzBuzzAI – The Most Inefficient FizzBuzz Solution Ever Made https://bit.ly/3HvbfPh May 29, 2025 at 12:19AM

Show HN: 3DGS implementation in Nvidia Warp: clean, minimal, runs on CPU and GPU https://bit.ly/3FneRCs

Show HN: 3DGS implementation in Nvidia Warp: clean, minimal, runs on CPU and GPU CPU & GPU with Zero Hassle Built with NVIDIA Warp, the same code runs seamlessly on both CPU and GPU — no need to deal with CUDA setup, driver issues, or device-specific kernels. Just flip one config line. Learn Modern Graphics the Easy Way Explore core concepts in differentiable rendering and parallel graphics programming — no need for expensive GPUs or thousands of lines of boilerplate. Minimalist & Educational This isn’t another massive codebase. It’s a clean, hackable implementation built for clarity — perfect for study, prototyping, or teaching yourself how Gaussian Splatting works. https://bit.ly/3FewJzs May 28, 2025 at 04:23AM

Show HN: AutoThink – Boosts local LLM performance by 43% with adaptive reasoning https://bit.ly/4dCoQAy

Show HN: AutoThink – Boosts local LLM performance by 43% with adaptive reasoning I built AutoThink, a technique that makes local LLMs reason more efficiently by adaptively allocating computational resources based on query complexity. The core idea: instead of giving every query the same "thinking time," classify queries as HIGH or LOW complexity and allocate thinking tokens accordingly. Complex reasoning gets 70-90% of tokens, simple queries get 20-40%. I also implemented steering vectors derived from Pivotal Token Search (originally from Microsoft's Phi-4 paper) that guide the model's reasoning patterns during generation. These vectors encourage behaviors like numerical accuracy, self-correction, and thorough exploration. Results on DeepSeek-R1-Distill-Qwen-1.5B: - GPQA-Diamond: 31.06% vs 21.72% baseline (+43% relative improvement) - MMLU-Pro: 26.38% vs 25.58% baseline - Uses fewer tokens than baseline approaches Works with any local reasoning model - DeepSeek, Qwen, custom fine-tuned models. No API dependencies. The technique builds on two things I developed: an adaptive classification framework that can learn new complexity categories without retraining, and an open source implementation of Pivotal Token Search. Technical paper: https://bit.ly/3Z31NJ2 Code and examples: https://bit.ly/3Sm2LMT... PTS implementation: https://bit.ly/4kbUxCZ I'm curious about your thoughts on adaptive resource allocation for AI reasoning. Have you tried similar approaches with your local models? May 28, 2025 at 03:39AM

Show HN: Weather2Geo – Geolocate screenshots from weather widgets https://bit.ly/45wVSzV

Show HN: Weather2Geo – Geolocate screenshots from weather widgets Hi HN, I built an OSINT tool called Weather2Geo that helps locate where a screenshot was taken based on the weather widget data visible (e.g. temp, condition, and local time in the taskbar). People often post these without realizing how specific those values can be in combination. It works by comparing the given weather condition, temperature, and time against current data from thousands of cities. It’s timezone-aware, supports fuzzy matching, and groups results geographically to reduce noise. It’s most effective on recent screenshots, ideally taken within the last hour—since weather and daylight conditions change quickly. The tool helps pinpoint likely locations when screenshots lack EXIF data or other traditional OSINT clues. It’s open source here: https://bit.ly/44WfzB2 Would love feedback, bug reports, or to hear if you think this has other use cases. - Elliott https://bit.ly/44WfzB2 May 27, 2025 at 11:31PM

Show HN: Millau – self-configuring ingress proxy for Docker Swarm https://bit.ly/4dAa9xS

Show HN: Millau – self-configuring ingress proxy for Docker Swarm I still use Docker Swarm. It's simple, fast to set up, and easy to teach. But I got tired of managing routing to microservices. Nginx and HAProxy need config redeploys, Traefik is powerful, but its configuration via labels quickly becomes unmaintainable. So I built Millau — a free ingress proxy and load balancer for Docker Swarm microservices. Add a few labels to the service, and traffic gets routed. No proxy restarts, no config files. Example: ``` # service deploy: labels: - "millau.enabled=true" - "millau.port=9000" ``` Millau listens to Docker events, discovers labeled services, and routes traffic using host and path matching. Unlike Traefik, Millau supports load balancing across services. I use it to deploy different versions e.g. Blue, Green, Red of the same microservice. If Red crashes, Millau routes to Blue or Green. If Blue slows down, Millau marks it inactive for 60s and routes to Green. Millau supports TLS termination and mTLS. It exposes Prometheus metrics and ships with a prebuilt Grafana dashboard. It's in production, serving its own site and several side projects. Let me know what you think, especially if you're still using Docker ecosystem in prod. Thanks. https://bit.ly/3Zq2gW3 May 26, 2025 at 10:15PM

Show HN: Get Red Carpet, CloseUp – Photoshoots in 10 Mins https://bit.ly/4dJ8NRy

Show HN: Get Red Carpet, CloseUp – Photoshoots in 10 Mins https://bit.ly/43nzCWo May 26, 2025 at 04:12PM

Show HN: JSON Commenter, add comments with valid JSON syntax https://bit.ly/4dxkHxA

Show HN: JSON Commenter, add comments with valid JSON syntax JSON Commenter is a vscode extension that lets you create inline comments in a JSON file while keeping valid syntax. You create a block with a command that places the comment anywhere legal in the JSON. The text is edited inline and supports word wrap, padding, margins, etc. The comment is in a somewhat good-looking box with minimal extra characters. Zero-width unicode chars give keys that only show two quotes. This is a sample comment block... " ":"----------------------", " ":" This is a comment. ", " ":"----------------------", https://bit.ly/4kkMRic May 26, 2025 at 11:33PM

Show HN: I built an AI image tool solo in 1 month https://bit.ly/4mBPEoI

Show HN: I built an AI image tool solo in 1 month Hey HN! Over the past month, I built Styleloop, a solo side project that uses AI to transform images into stylized artworks. The primary focus is image-to-image transformation, with curated styles like Ghibli, pixel art, Animal Crossing, and more. It’s my first real SaaS project, and honestly, the process changed how I think about building and shipping products. From backend to payments to UI polish — doing it all solo was exhausting and incredibly rewarding. Styleloop was built with Next.js, Supabase, Stripe, and Cloudflare. I plan to keep improving it and add more creative modes. I'd love your feedback on whether you’re into AI x creativity or solo hacking projects! Thanks for checking it out! https://bit.ly/45qsBqx May 26, 2025 at 04:12AM

Show HN: A minimalist web timer for focus and time tracking https://bit.ly/44NZxJo

Show HN: A minimalist web timer for focus and time tracking Started as a timer to track my freelance coding hours and stay focused. It’s local-only, zero-login, minimalist. Not monetizing it (yet), just seeing if others find it helpful. Thoughts welcome. https://bit.ly/3Sj7vTt May 26, 2025 at 09:16AM

Show HN : A noise free Hackers News newsletters + catch up page https://bit.ly/3FhMNAp

Show HN : A noise free Hackers News newsletters + catch up page Hey HN, I built HN500 because I wanted a no-fuss way to stay updated on major HN stories — especially after taking time off. It’s a newsletter and a catch-up tool with a few handy features: Set your own point threshold (250/500/750/1000) to get only the most upvoted posts. - Choose daily or weekly email updates. - Each email includes 3 random stories to help surface underrated gems. - The catch-up page lets you filter by points, time range, sort order, and more — great if you’ve been away for a while. It’s free, lightweight, and designed to stay out of your way. Feedback is welcome! https://bit.ly/4mwtbt1 May 25, 2025 at 09:17PM

Show HN: Manifold is a platform for workflow automation using AI assistants https://bit.ly/3ZnXeJF

Show HN: Manifold is a platform for workflow automation using AI assistants Hi HN! I'd like to show an early version of Manifold. This is a flow based UI (think ComfyUI) for designing automated workflows using language models. This is a passion hobby project and not intended for production use. The idea behind it was, can we create "the last tool"? A platform that can create any other computing tool or execute a process by designing workflows visually, and having AI assistants generate the necessary code or data with each run? Here is a list of some of the more notable features: - Supports public endpoints such as OpenAI API, Google and Anthropic completions, as well as any OpenAI API compatible local backend. - Integrated RAG solution that leverages PGVector for semantic search as well as an inverted index (keyword search) with multiple document merge modes. Option to retrieve full documents or chunks. - ReAct Agent with access to configured MCP servers. If no suitable tool is found, the agent is instructed to generate their own in a secure code sandbox. - Integrated web code editor that runs code in a WASM sandbox with support for JavaScript, Python or HTML previews. - CodeRunner node that supports third party libraries in a local container sandbox. Python, JS and Golang are supported. Libraries are automatically installed. - Nodes to control the execution flow of your workflow. Run concurrent agents. Jump to nodes. Execute for each loops on assistant outputs. Combine context. Publish and subscribe to topics to teleport context around your workflows. Documentation is non-existent but I am working on that before a formal 1.0 release. Expect bugs. Manifold is 99% a solo developed project with the help of a very much appreciated contributors. This is an open source project. Given the scope, I am sure I have overlooked many important things. Feel free to reach out to me if you have suggestions, or if there is a glaring issue that needs to be fixed. The README has multiple images of example workflows for your viewing pleasure. Hopefully those give the community an idea of what is possible when creatively combining the nodes. The repository is located at https://bit.ly/44UiNF4 . I appreciate any constructive feedback and contributions. https://bit.ly/44UiNF4 May 25, 2025 at 02:00AM

Show HN: I made a OSS alternative to Weights and Biases https://bit.ly/3FjuuLc

Show HN: I made a OSS alternative to Weights and Biases Hey HN! I made a completely open sourced alternative to Weights and Biases with ( insert cringe ) blazingly fast performance (yes we use rust and clickhouse) Weights and Biases is super unperformant, their logger blocks user code... logging should not be blocking, yet they got away with it. We do the right thing by being non blocking. Would love any thoughts / feedbacks / roasts etc https://bit.ly/4mxH6iw May 25, 2025 at 01:42AM

Show HN: I built StickerFacet to turn photos into high quality vinyl stickers https://bit.ly/43w45Sa

Show HN: I built StickerFacet to turn photos into high quality vinyl stickers I decided to focus on pets because my girlfriend and I have this cute little cat called Lola and I immediately needed to produce as many stickers of her as possible. Eager for feedback! Thank you https://bit.ly/4kt6T9G May 25, 2025 at 12:13AM

Show HN: I built a chill place online to work on your ideas https://bit.ly/4jkRA1M

Show HN: I built a chill place online to work on your ideas I used to watch hundreds of hours of lofi beats on youtube while I was coding, but I got really sick of all the ads. I decided to build a better alternative - It's a configurable space for you to hang out online while you work, with a ton of relaxing music and backgrounds. Its got useful tools built in like a timer to keep track of how long you've been locked in, and a notepad for todos or scribbling down ideas while you work. Honestly this is the first tool I've built that I personally use every day, so I'm hoping some of you out there can get some use out of it too! https://bit.ly/43zNb5j May 24, 2025 at 11:38PM

Show HN: HNRelevant – Add a "related" section to Hacker News https://bit.ly/3YZ9tMs

Show HN: HNRelevant – Add a "related" section to Hacker News It's been 2 years since the initial release [here]( https://bit.ly/43aGmG1 ). The initial version was a very basic prototype which was not available anywhere beyond the GitHub repo. You had to install it as a userscript or load the extension manually. Since then, it remained simple but better, new features include: - Improved accuracy by also using comments to help gauge the topic of discussion and the right keywords. - Published as plugin for more browsers: Chrome, Firefox (including android), and more recently Microsoft Edge. - Support for narrow screens and mobile devices. - Added preference controls. You're here because you love interesting HN discussions but they're often buried away like hidden gems, so give it a try and let me know what you think. https://bit.ly/432w2jf May 24, 2025 at 02:07AM

Show HN: High-resolution surface analysis with Lidar data https://bit.ly/43rYxIv

Show HN: High-resolution surface analysis with Lidar data https://bit.ly/3Fib3SY May 21, 2025 at 09:40PM

Show HN: Doc2dict a fast, open-source document to dict converter – No AI https://bit.ly/3FtnXNR

Show HN: Doc2dict a fast, open-source document to dict converter – No AI doc2dict is a python package that converts html and pdf documents into dictionaries preserving hierarchy. It also supports table extraction for html files. https://bit.ly/4kDf2sq Speed: * html - 500 pages per second single threaded. * pdf - 200 pages per second, pdf must have an underlying text structure. Multithreading is not possible due to the limitations of PDFium. Here's an example output from Microsoft's Annual Report: > "title": "PART I", "standardized_title": "parti", "class": "part", "contents": { "38": { "title": "ITEM 1. BUSINESS", "standardized_title": "item1", "class": "item", "contents": { "39": { "title": "GENERAL", "standardized_title": "", "class": "predicted header", "contents": { "40": { "title": "Embracing Our Future", "standardized_title": "", "class": "predicted header", "contents": { "41": { "text": "Microsoft is a technolo... Raw: https://bit.ly/3H9twBz... Parsed dictionary: https://bit.ly/4mshO5c... Simple description of algorithm: * Take complicated document such as pdf or html, and created a simplified representation for it as a list of a list of dicts where each dict is a text block with key features such as "bold", "font-size", etc and each line represents a new html block or line on a pdf. * Convert the simplified representation to a dictionary using a set of predetermined rules, e.g. smaller font-size for a heading means it should be nested under the larger font-size heading. Note that I am working on making the last part more modular by creating predetermined instructions that users can tweak for their use-case without rewriting the parser. I call these "mapping dicts". doc2dict also includes visualization tools for the debugging process: * visualize simplified representation https://bit.ly/43AmwFB... * visualize output dictionary https://bit.ly/43AmwFB... Why I made this: I'm currently working on another open source python package to make it easy to exploit Securities & Exchanges Commission data. Writing a generalized document parser that can be tweaked is easier than writing 100 or so specialized parsers for each document type. Also, converting html and pdf files to dictionary representation reduces document size by a factor of 10 or so. Not sure what I can do with that, but planning on some fun NoSQL database experiments. Link to other package (datamule) https://bit.ly/3SgjT6D May 23, 2025 at 10:01PM

Show HN: PolyLia – Learn languages by challenging friends (open beta) https://bit.ly/43JJY4o

Show HN: PolyLia – Learn languages by challenging friends (open beta) Hi HN! I built PolyLia, a multiplayer app that makes learning a new language more engaging by turning it into a game of real-time challenges. Unlike most language apps that focus on repetition or solo practice, PolyLia is about friendly competition — you challenge friends (or strangers) in quick language duels. The app is in open beta and available here: https://bit.ly/3YZtbro I’m looking for early feedback from language learners, educators, or anyone who enjoys gamified learning. What works? What doesn’t? Any ideas for features you'd like to see? Everyone who joins now gets a Founder's Badge! Thanks in advance — I’ll be in the comments all day to reply! https://bit.ly/4mzjPNe May 23, 2025 at 07:37AM

Show HN: Custom GPT to Discover Industrial Companies in the Benelux Region https://bit.ly/4kqxWT2

Show HN: Custom GPT to Discover Industrial Companies in the Benelux Region Working on a custom GPT to help profiles on the hardware industry in Benelux find the supplier/ service they need. It's particular to Benelux because in the region people speaks several languages and when you google in English you risk missing the info of websites in French, Dutch or German. I started with a knowledge based I curated based on info I had but activated the web search to find more. Feedback and tips to improve always welcome https://bit.ly/44PD8vl May 22, 2025 at 11:37PM

Show HN: Various ChatGPT Clients Written by Codex https://bit.ly/438BC6a

Show HN: Various ChatGPT Clients Written by Codex https://bit.ly/4mqQrZg May 23, 2025 at 12:50AM