496 results for working

news.ycombinator.com
Instance segmentation model that extracts 3D geometry from 2D floor planshttps://news.ycombinator.com/item?id=47092547Hey HN,I am an ML Engineer and a full-stack software engineer. For the past few weekends, I have been working on a pipeline to solve a PropTech problem: turning messy, highly occluded 2D floor plans into clean, structured data for 3D extrusion. Originally demoed for a firm hiring for the role.The Problem: If you try to use standard object detection (bounding boxes) or basic OCR (tested Qwen, DeepSeek) on architectural plans, it fails instantly. Walls intersect, doors swings and dimension lines heavily occlude the actual structures.The Stack & Architecture: I built an instance segmentation pipeline that relies strictly on pixel-perfect masking to pull the geometry.The Backbone: Swin Transformer + Detectron2.Model trained on 1024X1024 images, with RTX 4090Inference: Inference on CPU
Feb 20, 2026 7:17 PM
news.ycombinator.com
Show HN: AFS – filesystem-native memory layer for AI agentshttps://news.ycombinator.com/item?id=47062802I've been building multi-agent AI pipelines and kept running into the same structural problem: agents are stateless by default. Every session restart discards everything they learned. In multi-agent systems it compounds — Agent-1 learns something Agent-2 will never know. I started calling it "agent amnesia." AFS is my attempt to fix this. The central architectural decision is unusual: your filesystem IS the memory layer. There's no separate database process to run, no cloud service to authenticate against. AFS stores memories as JSON files in a `.afs/` directory, with SQLite FTS5 for full-text search, HNSW indices for vector similarity, and msgpack-encoded graph edges for relationships. *Three-tier memory lifecycle (automatic)* Memories auto-migrate without explicit management: - Working memory (< 24h): raw observations, fast access, no compression- Episodic memory: full history with provenance, searchable - Semantic memory: auto-consolidated knowledge (scheduler synthesizes patterns f
Feb 18, 2026 4:31 PM
github.com
Show HN: Corral – Auth and Stripe billing that AI coding agents can set uphttps://github.com/llama-farm/corralHey HN. I built Corral because every time I asked an AI coding agent to "add auth and payments," it hallucinated for an hour and produced broken code. Wrong imports, phantom endpoints, a login page wired to nothing.The problem isn't the agent. It's that auth-to-billing-to-gating is genuinely hard to wire, and there's no machine-readable spec for how to do it.Corral is an open-source CLI (MIT) that gives your agent a spec it can read (llms.txt), then scaffolds auth + Stripe billing into your existing project. It detects your framework (Express, Next.js, Hono, Fastify, and 13+ more), embeds into your existing server (doesn't create a new one), and generates working components: profile page, admin dashboard, plan gating, Stripe checkout, usage metering. One YAML config file controls everything.The agent workflow is 9 commands. Every command supports --json. Errors include a "fix" field. Exit 0 means deploy-ready.I hardened this across 10 framework/DB combos with blind agent tests: 3 AI mo
Feb 17, 2026 9:03 PM
coolwulfAI.com
Show HN: CoolWulf AI – A personal AI assistant built in Go, optimized for macOShttp://coolwulfAI.comHey HN,I've been building CoolWulf AI (https://coolwulfai.com), a self-hosted personal AI assistant. After seeing OpenClaw blow up, I wanted to share what I've been working on — a different approach to the same problem.*Why I built this:*I tried OpenClaw and found the Node.js/TypeScript stack heavy for what's essentially a local agent. pnpm, Node 22+, React — lots of moving parts. I wanted something that's a single binary, zero runtime dependencies, and feels native on macOS. So I built it in Go.*How it's different from OpenClaw:*- *Single binary, no runtime needed.* Download, set your API key, run. No Node.js, no pnpm, no build step. One ~100 MB binary. - *Built in Go.* Fast startup, low memory footprint, compiles to a native executable. No garbage collector pauses from a JS runtime sitting in the background. - *macOS-native integrations.* Deep AppleScript-based control of Apple Notes, Reminders, Calendar, Terminal.app, and WeChat desktop. These aren't browser hacks — they use the nat
Feb 17, 2026 5:32 AM
spacedaily.com
JUNO VR system brings detector events into immersive 3D spacehttps://www.spacedaily.com/reports/JUNO_VR_system_brings_detector_events_into_immersive_3D_space_999.htmlTokyo, Japan (SPX) Feb 13, 2026 Researchers working on the Jiangmen Underground Neutrino Observatory (JUNO) experiment have developed a Unity-based virtual reality system that immersively visualizes detector geometry and event information. The new framework is designed to overcome the limitations of traditional visualization tools and provide a three-dimensional, interactive view of complex neutrino detector data. The VR
Feb 13, 2026 5:20 AM
flipcloc.com
Show HN: An aesthetic customizable online flip clock and my first ever apphttps://flipcloc.com/I know there are many flip clocks out there, but flip clocks are truly wonderful and my version FlipCloc is smooth, optimized, detailed, and fully customizable to match your workflow / aesthetics / mood.Often, flip clocks online feel half baked: choppy animations, bad flipping animations, lazy flap separations, sometimes not even distinct top and bottom flaps, zero customization, and so on. After playing with time related concepts in various projects, I wanted to build an app of my own and made this using WebGL and React. This is my first ever app and I’m quite proud of this beta version.Try the full experience for free in any browser. No ads, just a beautiful clock you can run in a tab, window, or PiP while working or studying. (One-time purchase unlocks saving custom themes, settings, and presets and is usable on the web, app, or screensaver.) You can start by trying out some of the different preset styles and then customizing it to your preferences.Key features:• Show/hide seconds,
Feb 12, 2026 12:52 PM
spacedaily.com
ReOrbit and Google Cloud develop orbital space cloud networkhttps://www.spacedaily.com/reports/ReOrbit_and_Google_Cloud_develop_orbital_space_cloud_network_999.htmlLos Angeles CA (SPX) Feb 10, 2026 ReOrbit has launched Space Cloud, a multi year initiative to build a distributed network of software enabled satellites that can move and process data securely in orbit, in collaboration with Google Cloud. The company describes the project as a space based data center network that will combine on board computing, advanced networking and cloud style orchestration to support government, military a
Feb 10, 2026 7:40 AM
news.ycombinator.com
Ask HN: Where does modern geometry survive contact with SGD?https://news.ycombinator.com/item?id=46882060Over the past year I worked through Frankel’s “The Geometry of Physics” cover to cover, not to relearn physics, but to rebuild a modern geometric toolbox as it is actually used there: manifolds, differential forms, connections and curvature, Lie groups and algebras, fiber bundles, gauge structure, and variational principles.The motivation was practical rather than theoretical:Which of these geometric structures, if any, actually survive discretization, noise, and SGD-style training in modern machine learning?In physics, global and coordinate-free formulations were not aesthetic choices; they were forced when local reasoning stopped working. A recurring structural pattern was:structure -> symmetry -> invariance -> dynamics -> observablesIn modern ML we increasingly see analogous issues:* parameter symmetries and large quotient spaces * non-Euclidean data (graphs, meshes, manifolds) * highly structured hypothesis classes * training dynamics that are not well-described by flat Euclidean o
Feb 4, 2026 6:07 AM
news.ycombinator.com
Show HN: WarpParse – Rust ETL engine 1.5–8x faster than Vectorhttps://news.ycombinator.com/item?id=46880903Hey HN, We’ve been working on WarpParse, an open-source (Apache 2.0) ETL engine built in Rust, and we’re excited to share it with you today. We built it because we found that existing tools often struggled with resource efficiency or configuration complexity when handling massive log volumes. WarpParse aims to solve this with: Performance: 1.5x to 8x higher throughput than Vector in our benchmarks. Efficiency: Uses ~1/3 the CPU and ~37% of the memory of Vector under similar loads. DSLs: We created WPL (Warp Parse Language) for strong-typed parsing and OML (Object Modeling Language) for declarative data enrichment with native SQL integration. We’ve also built a visual editor (WpEditor) to make rule-writing less of a headache. We’d love to hear your thoughts on the architecture, the DSL approach, or any feedback you have after checking out the repo. GitHub: https://github.com/wp-labs Editor: https://editor.warpparse.ai As a fledgling open-source project, we deeply recognize the power of
Feb 4, 2026 2:59 AM
bing.com
10 Best Small Business CRM Software Of 2026http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a8a1baff9be441a8845106924f029db&url=https%3A%2F%2Fwww.forbes.com%2Fadvisor%2Fbusiness%2Fsoftware%2Fbest-crm-small-business%2F&c=267848304005587644&mkt=en-usShe has spent the last five years working on small business content to help entrepreneurs start and grow their businesses. She’s well-versed in the intricacies of LLC formation, business ta… CRM ...
Jan 29, 2026 9:36 AM
spacedaily.com
Sulfur ring molecule in galactic cloud links space chemistry to lifehttps://www.spacedaily.com/reports/Sulfur_ring_molecule_in_galactic_cloud_links_space_chemistry_to_life_999.htmlBerlin, Germany (SPX) Jan 26, 2026 Researchers at the Max Planck Institute for Extraterrestrial Physics, working with colleagues at the Centro de Astrobiologia in Spain, have identified the largest sulfur-bearing molecule yet seen in interstellar space. The compound, 2,5-cyclohexadiene-1-thione (C6H6S), was detected in the molecular cloud G+0.693-0.027 near the center of the Milky Way, about 27,000 light years from Earth. With a
Jan 26, 2026 6:55 AM
noriskillsets.dev
Show HN: A registry for curated, high quality Claude skills and skillsetshttps://noriskillsets.dev/Hi Hacker News!I’m Ritam, working with the small but mighty team at Nori. We’ve been obsessed in recent months with how to take tools like Claude Code from “I’ll experiment around with this” to “This is the most useful and necessary thing I use every day”. When I first sat down with our team to check out what they’d built, I found my skepticism about agentic coding melting away—they’d built useful, high quality, handwritten skills, instructions that functioned as “skillsets” to tie skills together for consistent and replicable results, and tooling to manage loading the right context for the right task into the agent.In recent weeks, the conversation around skills has reached a fever pitch, as have lists and sites full of skills scraped from all over the internet. Much like the actual gold rush, the current state of those collections requires a lot of time and sifting to find small, real chunks of 24k skills hidden amongst the muck (or to use a more relevant word, slop). So at Nori, we
Jan 22, 2026 5:00 PM
spacenews.com
Congressional hearing highlights military’s reliance on NOAA datahttps://spacenews.com/congressional-hearing-highlights-militarys-reliance-on-noaa-data/SAN FRANCISCO – A Jan. 13 hearing underscored the importance of ongoing collaboration between the National Oceanic and Atmospheric Administration (NOAA) and the U.S. armed services. U.S. Air Force and U.S. Navy witnesses discussed their heavy reliance on datasets and weather models provided by NOAA and said their agencies are working closely to ensure NOAA’s […] The post Congressional hearing highlights military’s reliance on NOAA data appeared first on SpaceNews.
Jan 14, 2026 12:41 AM
vercept.com
Show HN: Vy, a cross platform AI agent that automates apps without APIshttps://vercept.com/Hi HN, I am Kiana, I was an AI researcher for 10 years, and since last November I have been working on Vy. Vy is a desktop app for Mac and Windows. It sees your screen and controls mouse and keyboard to automate workflows across apps, including ones without APIs. I know there is a lot of skepticism about AI agents, and I share some of it. A lot of things in this space look impressive in demos and then fall apart on real, boring work. So here are a few concrete things that Vy does today that we and early users actually rely on: Where it works best Tasks that you could explain in a paragraph and that take a human a few minutes, not hours.Workflows that mostly involve clicking, typing, scrolling, and copying between a few known tools.Where it does not work well Very long, unsupervised runs. It is better to keep tasks bounded and keep an eye on it.Pixel precise work like detailed design or brush based editing.Some custom web editors or highly dynamic UIs still confuse it.How it is differen
Jan 7, 2026 2:32 AM
github.com
Show HN: WOLS – Open standard for mushroom cultivation trackinghttps://github.com/wemush/open-standardI built an open labeling standard for tracking mushroom specimens through their lifecycle (from spore/culture to harvest).v1.1 adds clonal generation tracking (distinct from filial/strain generations) and conforms to JSON-LD for interoperability with agricultural/scientific data systems.Spec (CC 4.0): https://wemush.com/open-standard/specification Client libraries (Apache 2.0): Python + CLI: pip install wols (also on GHCR) TypeScript/JS: npm install @wemush/wolsBackground: Mycology has fragmented data practices (misidentified species, inconsistent cultivation logs, no shared vocabulary for tracking genetics across generations). This is an attempt to fix that.Looking for feedback from anyone working with biological specimen tracking, agricultural data systems, or mycology.
Jan 5, 2026 6:30 PM
news.ycombinator.com
Show HN: A Multi-agent system where LLMs challenge each other's answershttps://news.ycombinator.com/item?id=46435835Hey HN,I've been experimenting with forcing multiple LLMs to critique each other before producing a final answer.In practice, I kept working around single-model limitations by opening multiple tabs, pasting the same question into different models, comparing responses, and then manually challenging each model with the others' arguments. (Maybe some of you can relate.) It worked sometimes, but it was cumbersome, slow, and hard to do systematically and efficiently.Based on my own experiments, a few things seem to drive why different models arrive at different responses: they have different guardrails, different tendencies encoded in their weights, and different training data. And the biggest kicker of all: they still hallucinate. The question I wanted to test was whether making those differences explicit, rather than relying on one model to self-correct could reduce blind spots and improve the overall quality of the answer.So I built Consilium9.The problem: When a single LLM answers a con
Dec 30, 2025 5:41 PM
thysys.com
Show HN: Calclock, Countdown/Lifeclock, Configurablehttps://thysys.com/happy-new-year.htmlI originally built this in Visual BASIC in 1994 for a friend who was in McMurdo Base Antarctica waiting on an 'Air Drop!' that was still months away, that would bring fresh food and much-desired mail to its denizens. It was a looming event everyone was anticipating and the waiting was cruel. No shortages but the 'freshies' run out first and for long spans everyone had to eat from cans, which saps morale. People would barter for spare batteries. I said via email "Sorry I cannot help you... but HERE is something that will make the wait even harder to bear! Hopefully it will become funny after awhile." So I sent him a CALCLOCK counting down to Air Drop. For awhile I used it to countdown to various things and to make life clocks.It stopped working a long ago. It was made using 16-bit Visual BASIC so ancient it even crashes the wine emulator today.This is the 2025 rewrite. Runs everywhere, 76k self contained html using js/svg elements built on the fly.Set from-time to NOW and to-time to the
Dec 25, 2025 12:39 AM
github.com
Show HN: Luxury Yacht, a Kubernetes management apphttps://github.com/luxury-yacht/appHello, all. Luxury Yacht is a desktop app for managing Kubernetes clusters that I've been working on for the past few months. It's available for macOS, Windows, and Linux. It's built with Wails v2 (https://wails.io), so huge thanks to Lea Anthony for that awesome project. Can't wait for Wails v3.This originally started as a personal project that I didn't intend to release. I know there are a number of other good apps in this space, but none of them work quite the way I want them to, so I decided to build one. Along the way it got good enough that I thought others might enjoy using it.Luxury Yacht is FOSS, and I have no intention of ever charging money for it. It's been a labor of love, a great learning opportunity, and an attempt to try to give something back to the FOSS community that has given me so much.A word of caution, in case you care about this sort of thing -- I wrote this with AI assistance. I am, first and foremost, a platform/infrastructure engineer, not a software develope
Dec 23, 2025 6:30 AM
spacedaily.com
Smart modeling framework targets 6G spectrum chaos in space air and ground networkshttps://www.spacedaily.com/reports/Smart_modeling_framework_targets_6G_spectrum_chaos_in_space_air_and_ground_networks_999.htmlTokyo, Japan (SPX) Dec 16, 2025 As wireless systems evolve from 5G to 6G, engineers are working toward networks that connect satellites in orbit, airborne platforms such as drones, and dense constellations of ground devices into a single space air ground integrated network or SAGIN. This convergence creates a complex electromagnetic environment where large numbers of signals overlap and interfere, making spectrum control and p
Dec 22, 2025 9:44 AM