Show HN for May 15, 2026

The core question: how did HP's scientific calculators actually work at the gate level? That rabbit hole led to building one from scratch.

The architectural decision everything else follows from: a decimal calculator should store numbers as BCD — one decimal digit per 4-bit nibble. A standard byte-oriented CPU (Z80, 6502) fights that layout constantly. So I designed a small custom CPU in Verilog where 4 bits is the natural data width and memory is nibble addressable.

What the project covers:

- Custom CPU: Harvard architecture, 12-bit ISA, 8-state execution FSM, hardware stack guard with a FAULT state for microcode debugging

- CORDIC for trig functions, verified to 14 significant digits

- Two-pass assembler in Python (~700 lines)

- Verilator + Qt framework: same Verilog source runs in simulation, as a desktop GUI debugger, as WebAssembly, and on real hardware

- Scripting language on top of the microcode for adding functions without touching hardware

- Custom PCB (EasyEDA/JLCPCB), battery, charging circuit

Write-up: https://baltazarstudios.com

Hackaday: https://hackaday.com/2026/05/13/build-the-cpu-then-build-the...

I'm a Type 1 diabetic and software engineer. Last year I went months between endocrinologists with no clinician reviewing my data. I'm an engineer, so I built the tool I needed — and now I'm open sourcing it. GlycemicGPT is a self-hosted platform that connects continuous glucose monitors, insulin pumps, and existing Nightscout instances to an AI analysis layer running on your own infrastructure. Data sources:

Dexcom G7 (cloud API) Tandem t:slim X2 and Mobi pumps (direct BLE) Nightscout (point it at your existing instance and you're running in minutes)

What the AI layer does:

Daily briefs summarizing overnight and 24-hour patterns Meal response analysis Conversational chat with RAG-backed clinical knowledge Predictive alerting with configurable thresholds and caregiver escalation

Important: this is monitoring and analysis only. GlycemicGPT does not deliver insulin, does not control your pump, and is not a closed-loop system. It reads your data and gives you insight on top of it. Your clinical decisions stay between you and your care team. Architecture:

Self-hosted via Docker or K8S — the GlycemicGPT stack runs entirely on your hardware BYOAI — bring your own AI provider. Use Ollama for fully local operation (no data leaves your hardware), or point it at Claude, OpenAI, or any OpenAI-compatible endpoint if you prefer a hosted model. Data flows directly from your instance to the provider you choose; nothing is routed through any centralized service operated by the project. GPL-3.0, no subscriptions, no vendor lock-in

Stack:

Backend API: FastAPI, Python 3.12, PostgreSQL 16, Redis 7 Web Dashboard: Next.js 15, React 19, Tailwind CSS, shadcn/ui AI Sidecar: TypeScript, Express, multi-provider proxy Android App: Kotlin, Jetpack Compose, BLE Wear OS: Kotlin, Wear Compose, Watch Face Push API Plugin SDK: Kotlin interfaces, capability-based, sandboxed

Looking for contributors — especially folks with BLE/Android experience or anyone in the diabetes tech space. Plugin SDK is documented if you want to add support for new devices. GitHub: https://github.com/GlycemicGPT/GlycemicGPT

Inspired by the recent Boards Of Canada announcement, I've been in a low-fi electronica mood lately and was going back and forth with Claude on how to design similar instruments in the browser that fit the genre. One thing led to another and pretty soon I had a fully browser based polyphonic synthesizer / drum machine / sequencer.

The interface and workflow was heavily inspired by the Rebirth338 application released back in the 90's, but with lo-fi synth voices rather than the original 303 & 808 emulation.

I know there's a significant overlap of developers and musicians and I though some of you may enjoy playing with the app, or at least listening to the resulting album. I've also open sourced track 1 of the album via the performance script used to record it. It's in the repo.

Bandcamp link to the resulting album: https://madmonk13.bandcamp.com/album/yesterdays-tomorrow-tod...

I used to study by rereading notes, and then I blanked in the exam hall.

Did some research, and found that my experience isn’t isolated, and that passive review doesn’t force retrieval, so nothing sticks, and I knew I had to do something about it.

That’s why I built Professor Goose. You pick a topic, explain it out loud to a goose, and he keeps probing until he understands you. Never gives you the answer, just keeps asking follow ups until a sound understanding is reached, which in turn makes you figure stuff out or realise you never understood your topic in the first place.

Free to try, no account needed, upload your syllabus for exam board specific questions. Curious whether this approach resonates with others, it sure has for me.

Hey folks,

I made a farm-to-door so you can find local farms near you.

It allows you to filter by farms that have raw dairy, unwashed eggs, and more “trendy” items.

It also lets you filter by who does delivery and pickup, so you can get your goods.

Most e-signature platforms charge: per seat, per envelope, API access fees, enterprise contracts

We built VisiSign with a different pricing model: $0.10 per envelope via API, no monthly minimum, no contracts, no seat pricing

Or $49/month for the hosted team platform: unlimited users, unlimited sends, API included

The reason this pricing works is simple:

E-signatures are infrastructure now.

Twilio doesn’t charge per employee. Stripe doesn’t charge per seat. Cloudflare doesn’t charge by company size.

But most e-signature platforms are still priced like 2000s enterprise SaaS.

It should work more like payments or email infrastructure: API-first, automation-friendly, embedded into workflows, usage-based when appropriate

Real signature infrastructure is more than rendering a PDF: audit trails, timestamps, signer event history, tamper evidence, completion certificates, webhook workflows

So we built VisiSign around that assumption.

https://visisign.app

A lightweight, keyboard-driven Kubernetes TUI for exploring and managing clusters from terminal.

I was tired of paying $100s/mo to access data I should own -- my own DMs on social media -- so I built Allman, a local-first cli to access linkedin messenger.

Starting with LinkedIn, I gave the entire compiled js binary of linkedin's web app to claudecode and reversed engineered the entire messenger inbox in 24 hours. My goal is to bring this to all messengers so AI can handle all of this busywork, just like it can my email.

I had also never built a TUI so I had claudecode build me one based on top of the local CLI and filesystem.

Full repos:

https://github.com/tarkaai/allman-cli

https://github.com/tarkaai/allman-tui

Their sharding / access is brutal, so it'll likely break as-is -- but the point is that Reverse Engineering is trivial now (and can do it dynamically through browser access and dynamic playwright -- thanks Browserbase! ).

Would love to hear your feedback and what I should build on it next.

Shoutout to Eric Allman.

Hi HN,

I'm a longtime web dev, but this is my first macOS app. Feedback is very welcome <3

I built TongueType because every dictation app I tried failed at least one of three things: it sent my audio to someone's cloud, it charged me monthly fee, or it felt clunky or enterprisy. The cloud thing kills it for anything sensitive. The subscription thing is insulting for what amounts to "Whisper + a hotkey."

TongueType runs Whisper locally on Apple Silicon via CoreML. It's built to feel like a second keyboard: tap the hotkey, talk, let go, words appear. It sits in the menu bar (no dock icon, no buttons to click). I use it constantly for LLM prompts, code comments, emails, and DMs. It gets out of the way completely. (It can also transcribe audio and video files.)

No accounts, no servers, zero telemetry. The free tier is 30 minutes of live dictation per month. Pro is $19.99 one-time for up to 5 Macs. No subscription, and I'm committing to keeping it that way.

I wanted the app to be fun and personal. I included twenty accent colors including Rainbow Mode that runs a satisfying gradient through a waveform while you talk. Custom listening labels, adjustable overlay position, spoken cues like "new paragraph" or "scratch that." And customizable post-processing rules! I want the app to feel like its yours instead of something you're renting.

Two things worth mentioning about what didn't work:

First, I tried building an iOS keyboard extension as part of the app. iOS dictation just isn't great, and a local Whisper keyboard would have been a real upgrade. Turns out keyboard extensions can't access the microphone API at all on iOS, so the the only workaround is bouncing out to a host app to record then back to the keyboard, which is a terrible experience. Because of that, it's macOS only.

Second, I originally submitted it to the Mac App Store and it was rejected under Guideline 2.4.5, which says an app can't use Accessibility APIs for non-accessibility purposes. This is the infamous rule that gets most dictation apps rejected. The irony is that third-party dictation apps have a very real accessibility angle. People with RSI, tremor, arthritis, post-stroke hemiparesis, etc. often find long-form typing to be painful or even impossible. Apple's stance is that "inserting text at the cursor" isn't an accessibility use case, even when the user literally can't use their hands. So it's a direct download, signed and notarized, outside the App Store.

And a huge shout out to what worked really well: Polar (polar.sh) was an absolute joy to setup and use for payments! It feels like Stripe circa 2011. It's simple, fast, intuitive, and just works.

Website and download: https://tonguetype.app/

(My daughter did the voiceover for the video...she did a great job!)

I'm happy to answer any questions, and genuinely want to hear what's broken or could be better!

With coding agents now allowing for lots of small projects and products running in tandem, and as they tend to grow in operational complexity, I've been finding myself wanting a way to manage a bunch of running npm/bun servers as well as one-off operational jobs I need to run. I also want to keep my terminal focused on git and claude, and have my servers somewhere else.

Nemo solves that problem: I can now more efficiently manage and view local servers and jobs, look at their logs and see their CPU/RAM metrics as well. My terminal windows are now much more tidy and focused.

It's not the most complicated project, but posting it here in case it's useful to anyone else.

I make fractal art with GLSL and got tired of fighting Shadertoy's limits when I wanted to actually finish and export pieces. So I built the tool I wanted: ShaderKit runs in the browser, fully ShaderToy-compatible, but adds the export pipeline that was missing for me.

- Loop examination mode: visualizes the transition point of a looping shader in slow motion so you can see exactly where it breaks. X-time remapping helps debug discontinuities by mapping time to the x coordinate, this way you can see literally if the loop is seamless or not, if you see any seams, its not seamless. This is what I needed most for my own work.

- Cloud rendering up to 8K with supersampling

- A GIF exporter allowing for setting custom color palettes and choice of dithering algorithm

The main editor and features are free. Cloud rendering is paid because GPU time costs money. There's an optional Claude-based vibe-mode as well.

Would especially love feedback on the loop tools from anyone who writes seamless shader loops, that's the part of the problem I think is most under-served.

https://shaderkit.com

I’ve had this idea for a new code editor chord grammar bouncing around in my head for over 2 years, and now I can finally release v0.1 of ane. You can run it one-shot from scripts or agents (`ane exec —chord…`) or use its full-on TUI editor. ane is pure Rust, and integrates tree-sitter and LSP to provide a pleasant and predictable terminal code editing experience. I’d love folks to try it out and give feedback! Thanks :)

X open sourced their algorithm so i read through it and used claude code to write a plain english breakdown of all steps

Problem

A few years ago I got my first ultra-wide monitor. It was a 49" Samsung monitor. I work as a software engineer, and in my line of work I need to screen share a lot. This almost immediately became a problem due to the size of the monitor. I would either need to get a second display just for screen sharing, share one window at a time (which required me to constantly stop/start sharing to select the new window), or share my entire screen making it difficult for other participants in the call to see what I'm doing due to the size of my display.

I wanted a solution to this. At the time, I was on Windows. So, I wrote a very basic application and released it for free on GitHub. This was called "FrameCast". It had a lot of issues, including usability issues. But at the time, it got the job done. After a while I moved away from the ultra wide monitor, so I stopped using the application.

Fast forward to 2026 when I again purchased a new Ultra Wide monitor. This time, using macOS. I ran into the same problem again, and when I loaded up my old FrameCast application, it no longer functioned properly on macOS.

This led me to creating Raybeam.

What is Raybeam

Raybeam is a native macOS menu bar application that aims to make screen sharing more accessible for users.

It features:

    A draggable, resizable region of your screen that can be shared in video conferencing applications.

    The ability to filter out specific applications from being included in the shared region (this is useful for keeping things like Messages or Discord out of the shared feed, even when they are visible on your screen)

        I believe this is one of the key features that sets this application apart from other similar applications.

    The ability to notate the shared region with hand drawn notation.

    Customizable hot-keys for quickly changing what region of the screen is being shared, toggling the live feed and quickly toggling "draw" mode.

Compare

I'm only aware of one similar application to this, which is AnyFrame. Having purchased and tested AnyFrame, it seems to be very laggy when dragging the overlay around, doesn't offer anything similar to Raybeams "hidden apps" feature, and has a much higher barrier to entry with a $70 lifetime license. None of these are issues with Raybeam.

Pricing

I've tried to remain fair with the pricing on this application. I see far too many applications being released with subscription based pricing models that I don't personally believe are reasonable considering the application that's being offered. I didn't want to do that with Raybeam.

This application is priced at $9.99 for a lifetime license on the Apple App Store.

I had a bunch of custom AI pipelines and a growing folder of markdown files and Python scripts holding it together. Built this to give that chaos some structure.

Agents are markdown files, topology is a JSON file the runtime enforces hard. The agents are still fully autonomous: they make their own decisions, but the graph they operate in isn't. You declare who can call whom upfront and the runtime holds that line.

No auth yet, fine if you don't expose the port, i guess. Two Docker commands to run it.

My latest project, about 60% of the codebase was written with Z.ai's GLM-5.1 model. It's basically a Telegram bot that allows for embedding/downloading media easier within group chats. Absolutely 0 telemetry and you're free to edit the code as you wish. I know this is so small and stupid compared to the cool stuff I read here every day, but I'm very proud of my accomplishments as I had to learn a lot to get this to work as intended. Anyways enjoy ^^

I built Blogr because I've shipped a handful of side projects and can never stick to blogging, even though I know it's the best thing to do for SEO long term. I want to spend my time building, not blogging!

Blogr is a GitHub App. You install it on your repo, configure your site once (description, tone, audience, keywords you want to rank for, topics to avoid), pick a schedule, and it commits MDX blog posts to your repo automatically.

The positioning for this app over others like it:

- Posts commit as MDX files in your repo, so you keep them if you cancel.

- It reads your existing posts to match voice, avoid duplicate topics, and add internal linking.

- If you're on something like Vercel, posts auto-deploy as soon as Blogr commits.

- Posts are run through multiple filters to maximize SEO and humanize the writing.

- Set and forget. Nothing to monitor or babysit.

Happy to answer anything about the implementation or the idea.

When landing data I prefer to keep it as close to the original source as possible. Most of the Python data ingestion programs I saw treated Python more like SQL instead of harnessing object orientation. This was my attempt at translating my object-orented columnar approach to Python. I originally did it with Requests and Pandas but the overhead costs did not seem worth it. Claude helped refactor for async and Pydantic.

Now HTTPX’s async capabilities and Pydantic’s class building took this project over the top. By harnessing their abilities I shifted the codebase from data mapper to pipeline orchestrator. I added every format I could that seemed to have an established Python library. Right now I believe I support : JSON, NDJSON, XML, CSV, TSV, PSV, SQLite, and HTML out of the box. Optional extras (~30 MB pyarrow) unlock Parquet, Feather, ORC; Avro and XLSX have their own extras. I also added every compression I could find. Benchmarks at least for a windows machine are on par with other elt packages.

By focusing on function wrappers to make the developer’s syntax as easy as possible for the original data mapping calls, I established simple automated pipelines with one cli command and one JSON reference file. The JSON is basically the same syntax you would use in Python.

Both stream and fjord accept inflow and outflow Python code. Inflow code allows you to set custom conversion functions and mappings for the incoming data. The outflow code allows you to manipulate the exporting data into a new object new entirely.

Also, because your pipeline is basically created by a JSON file. You should eventually be able to automate the creation of the entire pipeline. Enjoy.

https://github.com/PyPlumber/Incorporator/

How you use it: Declare a subclass with no fields, point it at a URL, and it infers a Pydantic model from the response at runtime — with full strict typing, dot-notation, and an optional registry lookup by any key. class Launch(Incorporator): pass launches = await Launch.incorp( inc_url="https://ll.thespacedevs.com/2.2.0/launch/upcoming/" )

These functions handle the rest of your data mapping and export format needs: - test() lets the framework write the call kwargs for you - refresh() re-fetches with the seed call's params auto-replayed - export() serialises to any of the 13 formats

Then these functions create a pipeline. - stream() runs a chunked daemon with bounded memory. Can be used in two modes: pass-through or stateful (in RAM) updates to be manipulated in real-time. - fjord() fans out N sources and fuses them through a user reducer. This accepts multiple sources and exports.

After that all works copy the parameters into pipeline.json and the command can be as simple: incorporator validate pipeline.json incorporator fjord pipeline.json –logs

Same interface as tee. Writes to files you don't own.

Hey HN, Aron here, founder of Gatsby. Today (May 14, 2026) we made U.S. history with the first humanoid robot cleaning service delivered to a consumer in the US — we picked someone off our SF waitlist, drove a humanoid to their apartment, and it cleaned the place.

Pricing is $150 flat in SF, any size apartment. Cleaning took a few hours including setup.

Curious to hear any thoughts and happy to answer any questions.

anything from "uk garage tropical house" to "music for fixing a car at 2am". type a mood, get an endless custom radio station. multilingual

Hey HN,

Founder here. I recently left a job and had a lot of time on my hands so I decided to build an app that will help people navigate the current job market and ensure they are being paid fairly.

I wanted to build something that gives the career benefits of LinkedIn without the social feed or self-promotion. Let people privately post about their work, keep track of their brags, and have an idea of where their career gaps are. Maybe help them to realize they are being underpaid so they can make a bit more.

Applied to YC on a whim but I'm thinking that's more of a lottery ticket. Mostly hoping to help people, build something worthwhile, and make enough on the side to cover hosting costs.

We have public profiles too for people who want to show off their work but not deal with social engagement. Here's mine: https://proving.app/@jondaniel.

Feedback is welcome. Thanks everyone!