alt Hacker NewsAsk HN: What are you working on? (November 2026)
What are you working on? What have you been curious about lately?
Comments
>Making a simple game
>Want player scripts to be possible (And dogfood it for enemy ai too)
>Behavior tree dep looks too easy, let's make a dsl + interpreter
>Typed builders are annoying (because stateful), lets separate into builders -> untyped ast -> typed ir
>Hm, typed ir is super repetitive, now I have a schema (ts) -> codegen (ts) -> .ts / .rs build step
Fun waste of 2 weeks, not even counting the time spent on pl literature
A suite of own cloud based local first apps optimized for eink tablets, specifically the boox palma 2. Notes, book reader, browser, music player, maps, camera etc. With a companion ios app combining all of them.
this way when i want to disconnect i can just grab the palma and go
I've been working on an Obsidian plugin to quickly enter caloric and nutritional data (https://www.github.com/Krishna-Sivakumar/joule). The core features are done (A DSL and a runtime for finding and resolving formulae recursively), and I need to work on documentation and some convenience features. Hoping to finish it this month!
A domain search app called NameTheory: https://apps.apple.com/us/app/nametheory-domain-search/id680...
It uses Apple Intelligence to help brainstorm domain name ideas. I’m interested in seeing what small local models can do as they catch up to larger ones, this is my first iOS app so I started with something simple that scratches my own itch.
Maintaining my indie social network for tech people https://stackness.dev/ (currently struggling to keep users on the platform, even games and achievements didn't help, can you believe that? :D)
and this weekend I'll finally launch a tool for guitarists and their friends https://kantamigo.com/
I'm reverse engineering the streaming protocol for three different brands of cheap surveillance video cameras. Each brand has its own app. They're all awful. None of them have a native desktop MacOS app. I've successfully created a native app for MacOS and Android for one of the brands. They're faster and work better than the ones they're replacing. One down, two to go.
Working on a natural language intelligence layer for contacts - "Pull everyone I know who works at a Series A healthtech in the Bay Area who also went to Stanford for ungrad" => either structured or unstructured results. Large-scale data support, automatically enriched with up-to-date data, fast/accurate results. Integrates natively with Gmail and Gcal.
Looking to pilot for free with small qualified teams (early-stage startups with real/immediate need).
When I was a kid I loved a game called "Stars!" from 1995. I cloned the core and I'm also adding my own touch and lots of high sci-fi concepts. Already fun to play, early access coming soon: https://store.steampowered.com/app/5482440/Asteria_4X
I've been working on a voxel game engine called Bonsai for ~10 years.
Probably the most interesting thing about it at the moment is the editor. The world, and most things in it, are represented as collections of SDFs. More accurately, they're density fields, but, potato-tomato. One neat thing about representing the world as SDFs is they've been thoroughly researched and documented by a guy named Inigo Quilez, and they have a lot of nice mathematical properties. For example, you can do a smooth union of arbitrary SDFs to get nice rounded contours where shapes join.
The world generator and editor operate almost entirely on the GPU. The terrain and editor pipeline is a three stage process. The first stage is user-configurable; users write an arbitrary 'shaping' shader in GLSL. The second stage is a fused stage; it does a convolution on the output of the first stage to compute a normal, runs a second user-configurable 'decoration' function, and finally applies edits. Edits are collections of SDFs that are 'compiled' from a graph structure on the CPU into a bytecode, and interpreted in the shader. The last stage runs another convolution to compute the final normal, then packs an occupancy bit, color, and normal down into a 32-bit representation, suitable for transfer back to the CPU. I expect I will expand this to include additional user-level bits for encoding biome type and GPU-driven spawning.
The entire shader pipeline is one fused 'ubershader' .. everything is bound at startup and the shader is dispatched three times, incrementing a 'Stage' uniform on each dispatch. This strategy conceptually makes the driver overhead nearly zero, and the profile agrees.
The final step in the process is giving the CPU a chance to spawn additional entities and structures after generation. This is not completely fleshed out, but the bones are there. The idea is that the results of the GPU work are passed back to the CPU, which can run arbitrary callbacks to spawn additional edits and entities. Once spawning is complete, the result is meshed using a fast binary meshing algorithm and uploaded to the GPU. I keep 16 bits of data per voxel from the GPU work, 1 bit of occupancy, and 15 bits of normal for collision response, which works out to half a megabyte per chunk. At the moment I only keep chunks which contain a piece of the world surface, and discard the rest. Surface chunks are dense, and may be compressed in the future, but keeping this amount of data scales to world sizes greater than 1B voxels cubed, which seems good enough for now.
In terms of raw performance, GPUs are ridiculous. My 3080ti mobile gets somewhere on the order of 500B-1T perlin samples per second with a naive perlin implementation. Compared to FastNoise2, that's about 1000x more, which sounds wrong since my GPU only has 35x more capacity, but, that's what my initial profiling said. That number is the marketing number; the real throughput is significantly less due to multi-stage nature of the system, but, there's a lot of headroom for improvement. For example, dropping the backing textures from 32bit to 16bit roughly quadruples performance. Since we only really care about the surface where values are near zero, I believe using 8 or 16 bit is a viable option, though this claim will require a fair amount of study to validate.
All in all, I'm pleased with how this is all working out. My original goal was to write a highly flexible system giving users both easy-to-use tools and granular control over the final output. I think I've done a fairly good job at that, and the fact it's also super fast is the cherry on top.
I'll do some shameless self-promotion and leave some links here for anyone interested in looking at the engine, language code, or some pretty pictures.
A video game in which players can build their own worlds. The brains of the creatures in the video game are a small language model + harness that runs in your browser: http://id-le.com
It’s open source: https://github.com/nicobrenner/id-le
I'm still working on my own shell based AI tool. This month I am testing and reworking it to try and get useful results with really small models, like 2B level. Goal is to get help debugging Linux issues when offline with these small models that can run anywhere. Then also being able to access cloud based models to do harder stuff once online. all while feeling like a CLI tool.
github.com/patrickjh/ssa
I've been building data pipelines from sometime as part of some other products, eventually pulled out a part of that system into a standalone product called hydrafetch.com
That's the main thing that I'm working on these days
We've been reverse engineering the tensor core on TPU v5: https://gherkinlabs.com/v5e-tensorcore-instructions
The work mainly stems from a belief that Nvidia gpus' architecture are a bit subpar for machine learning and that tpu could be a viable alternative but are still rather undocumented.
Working on a tool to provide a thin OIDC auth layer for self-hosted LLM engines. Users log in with your Org's identity provider, generate an API key for themselves, point their harnesses at your instance of the relay, and then they're off to the races.
No web chat interface, no call-us-for-pricing feature gating, just a tight Erlang program for people who want to get shit done!
I’ve been working on my inertial navigation library. It actually beats ArduPilot’s navigation subsystem in my benchmarks. I’ve been working on these algorithms since my PhD thesis and now I am releasing the stuff as open source embedded C and Python library:
I'm building mensara, multi player memory so you're whole team can work together with ai coding software.
One of the most fun things is you can use all the various agents, like muse, grokbot, dots etc and they all know you immediately and can work together.
Try it at www.mensara.ai
A cli command to view PDFs, images, Word documents, spreadsheets, drawings, Markdown and HTML inside the terminal (kitty, ghostty, etc.)
Works well with tmux and through ssh too:
I've been working on a small search engine for content creators that play and/or review indie games. It's written in C and has been a lot of fun to work on.
Still building the file manager I've always wanted: https://intestinate.com/gusco/
Added an audio player this past month for all fellow file hoarders out there.
Also, you can now choose if you want Gusco to live in your terminal or run as a native-yet-texty app.
Was pleasantly surprised how well turning a Bubble Tea TUI app into a GUI-ish bundle works with Gio UI.
Answering the second question:
I've been curious about...
The machinations of my own mind/psyche/nervous system. For various reasons, I have had to get increasingly curious and stay that way over a years-long haul. This includes the usual suspects:
- "Serious Stuff", lets say. Viz. paying attention to and learning about therapeutic introspection, relationship/social dynamics, clinical psychology, meds as they apply to me.
- BUT also can include "fun stuff"! For example: seeking out and befriending like-oriented kindred spirits (it feels great to be seen / heard by peers), and developing a persistent interest in biology (which I hated in school), and even signing up for research studies! I got accepted into a Phd memory study and it was great to be back on a university campus, a couple of decades since graduating.
Current opinion: I think all mental health work leans N(1).
The art of it appears to be remaining curious and seeking self-understanding without lapsing into tiktok pop-psych, nor a false sense of studied erudition (No; one will almost certainly not become a clinician).
The greater the complexity/depth/intensity of one's condition, the more it calls for subtle, bespoke work. Progress from outside-self intervention has an upper limit (e.g. "taking the prescribed meds"). Active, self-directed self-work (in concert with one's support system, of course), brutal as it, must be done... somehow.
To others in a similar state space; may The Force be with you. Following one's curiosity, even now, will probably not result in the curious cat's fate. Worth the extra (omg so much extra) trouble, IMHO.
_|/
P.S. Anyone wishing to discuss more, feel free to email me (see bio). Please refer to this exact comment.
I've deliberately not shared specifics, including self-educational resources, because this stuff is slow, plodding, deliberate N(1) work, and I've personally found it counterproductive to read/consume the vast majority of N(1) experiences I've found.
(edit: fix typos, reformat for clarity)
Vicena.ai allows you to design, simulate and test new battery chemistries, new drugs… With the same ease as building software with coding agents.
It is an AI for Science connected to scientific compute and a network of labs, so you can do science end-to-end from your computer.
Infrastructure for hosting static webpages with options for storing data, having users, sending notifications, shared "rooms", and various other bits and bobs all behind an mCP. All $0 practically because it's made up of static services from AWS.
Made for me, but I can provide access to friends and family, which is nice.
I just celebrated the 1 year anniversary of Pixel Brite, the media server Plex never became.
I started a company to focus on the niche accounting professional service called Cost Segregation in the summer of 2024. I was laid off from Meta in February 2025 and dove in full-time on that startup, Segtax.
We are on the cusp of automating the bulk of the work. It's been an exciting ride
My integrated terminal + file browser with enhanced autocomplete and search. Replaces all native OS ones for me: https://github.com/David-OConnor/peregrine . Also includes misc utility OS functionality, tailored. (Advanced clipboard, a scratchpad etc)
Building an intelligent typing application. https://typequicker.com
Anyone can learn to type +100wpm - just have to focus on your weak points and spend 5-10 minutes a day for a few months.
A modern take on the game FTL: Faster Than Light, with inventory autobattler gameplay. It is called Cosmo Cargo and a demo is available on Steam: https://store.steampowered.com/app/4659480/Cosmo_Cargo_Demo/
https://petrosprawl.fyi/ An attempt to map oil infrastructure around the world
https://chord-network.vercel.app/ A tool that helps you build chord progressions
Ive been blogging about learning how LLM services work. Just finished the 2nd part and hoping to do the last part this week. https://matthewtejo.substack.com/p/designing-an-llm-chat-ser...
I've just released v1.0 of Rouler which is VPN split tunneling for windows. Think IP rules for Linux, but with per-app/service connection steering. The whole app is built in Rust, including the kernel driver. Having released v1 I'm now starting work on adding folder and windows store support. https://rouler.io
https://github.com/Polign/recall
Polign Recall, a state and memory management for agents that can provide append only state logging and governance. The memory and storage is completely on BYOC setup and can work with confidential workloads
I’ve been working on an accounting agent factory (https://www.translucent.io) . Building a harness and infra that can process books with thousands of invoices, transactions, etc. is quite rewarding and challenging. In particular the long-running nature of things.
I recently made a mobile app for practicing singing: https://singtrace.karamagic.com
The free version you can record and keep an archive of your daily practices of each song, while a paid version does some analysis of your vocals.
A free comedy newsletter that has nothing to do with technology yet mentions it from time to time.
Still working on ShopSpec, a parametric deterministic furniture builder. Trying to find time to put more actual builds in the gallery - have a few projects that are in progress while continuing to build out the tool and iterate where I find personally useful.
Started making youtube videos again, using my human brain to pick the topics and write the scripts, and AI to edit them (mostly).
Here's one about speculative decoding (and how it speeds up AI inference): https://www.youtube.com/watch?v=cGnIKjGIwjA
I built Telemetry Machine (https://telemetrymachine.com).
The TLDR is that if you like Honeycomb or Datadog but don't want to pay insane prices, you should check it out.
The longer story is that I love talking to SREs, platform engineers, etc about observability, so if you want to kvetch about it, please send me an email at [email protected]!
My cofounder and I have been working on observability for a combined 15 years and we decided to take what we love most about all these tools. We've got the powerful querying of Honeycomb, the great UI of Datadog (though most users use us via agent anyway), and the speed of Clickhouse.
Check us out!
I have spent the last six months or so on a multiplayer, open world game. The idea is you can jump to any city, walk around, see businesses and chat with others. Loading the resources (map etc.) as they are needed and garbage collect have been a challenge,
Building a soccer game based on Mike Tyson's Punch Out, just to see if I can create some method to see if an agent can iterate on making something more fun. Obviously that depends on the definition of fun (which is already a can of worms), but it's a fun way to use a different side of the brain.
https://hn.fiodorov.es AI-powered search across all HackerNews comments
https://climate.fiodorov.es Agentic analysis of climates of world cities with more than 300k inhabitants
Both have mcp.
For the month of November (and starting in October this year) I like to hold weekly cooking challenges to practice and refine my Thanksgiving day feast. Now about 3 weeks out from the final boss level of cooking, and I'm having a week all about brussel sprouts. Earlier weeks were cauliflower and the classic seasoning blend of sage, rosemary and thyme.
A few projects, but now waiting for a week for my Piano Game[0] to be approved on Play Store. Maybe it takes longer cause I marked it as being "for kids" too.
[0]: https://ultimidi.com
I'm working on a data monitoring company:
https://turbolytics.io/blog/debezium-wal-growth-postgres
Happy to get your postgres and kafka connect monitored and alerted on for free!
I've been updating the landing page for my book https://networksbyanalogy.com website to be dynamic so users can "explore" the book in a visual way.
Sysadmin agents on old machines.
Run AI on machines 30 years old.
Entirely in legacy Perl.
HP-UX, OpenStep, Solaris 8, Tru64, Irix, old Linux. Legacy FreeBSD, NetBSD...
Optional desktop in tk.
There's even a proxy for modern tls.
I've been working on https://gitsocial.org/ for a year and a half now. The approach is to keep it all in git: issues, PRs, comments, releases, etc.
Hoping it becomes a viable self-hosted alternative to GitHub, GitLab, and Codeberg!
I have been working on a derivation step auditor for checking any mistakes in your (or your AI buddy's) mathematical derivations!
Building a native macos app with a suite of completely local AI tools. The first version is out and have a long roadmap in front of me. Would love to hear some feedback.
I'm building Oculette (https://prelaunch.oculette.ai/).
Basically, I hooked up an ESP32 camera to a vision model so I can prompt it to watch stuff and send webhooks. Thinking analog gauges, legacy equipment, sourdough starters (still no API), and stuff like that.
Been dogfooding the prototype on my desk, and now I have an air quality API that reads the display on a cheap monitor.
Also added direct camera access via MCP so Claude could help me debug the hardware, which has been pretty fun.
Turning RSS, atom, blog feeds into audio feeds with kokoro tts. Free, web app only.
https://porchlit.app/
I built the original for myself because I wanted to listen to acoup and other history blogs while sleeping. And in the spirit of getting better at finishing projects decided make it public even tho it’s nothing really unique or special.
Still lots to polish especially in the text/copy and ui but the core features should be working.