alt Hacker NewsAsk HN: What Are You Working On? (December 2025)
What are you working on? Any new ideas that you're thinking about?
Comments
We want to speed up adoption of custom AI, but most people suck at building it (no expertise, money, time, etc.).
We thought, what if you could "Vibe ML" your way to it? Allow any AI engineer or PM to build custom AI directly from their current implementation.
So we built these agents that orchestrate the entire life-cycle of custom AI. We start by hooking into how you use AI, prepare/label your data, detect the best recipes for your task, fine-tune, and deploy it for you. Really tried to simplify the entire process.
We aren't entirely sure about the UX/UI patterns. We aren't going chat first because if most people don't know where to start with ML, how in the world are they going to prompt it!?! Instead, we auto detect the AI tasks you've built and go from there.
An AI assistant plugin for Logseq. https://github.com/shovon/logseq-ai
It allows users to "chat" with their Logseq graph. Think of it like a "Cursor for Logseq". I hope people find it useful. I have on numerous occasion wished that I could have easily asked about a specific block on my graph, and would provide an intelligent response, also somewhat influenced by the contents of the entire graph. It's still a work in progress.
It's fully open source.
Building https://localhero.ai, automated on-brand i18n translations that run in your CI pipeline. Right now I'm working on better .po/gettext support, based on feedback from an early customer. With gettext you usually keeps source strings in the actual source code. So I'm building a workflow where non-technical people (PMs, designers) can edit translations in the web UI and then easily generate a PR with both code changes and translation file updates. Trying to make translations work smooth for both automated CI pipelines and PMs/designers who don't live in Git, when translations are checked into the repo. Also going through my network, talking to devs and localization folks to understand what could be improved in their orgs for translations.
Still working on the Mint programming language (https://mint-lang.com/) with a 1.0 release in January :). I'm happy with the current feature set, so I'm just polishing and optimizing where I can and giving the documentation a throughout look.
I got a couple new toys for birthday/xmas: the GPD MicroPC 2 UMPC and the M5Stack Cardputer.
The MicroPC is great because it makes it super easy to code and hack on something in places where it would be too awkward or annoying to whip out my laptop, and the Cardputer is just a fun little toy because it's so open ended and hackable. I've been writing an app for Cardputer to control my thermostat remotely, and I've had a lot of fun grossly overengineering the needless amount of concurrency I have added through FreeRTOS.
Something oddly satisfying about using a micro PC to program an "even more micro" PC. What a cool time to be alive; I would have killed for this kind of stuff as a teenager!
Publishing everything local councils do in the UK at https://opencouncil.network - trying to help people feel like they know who and what they’re voting for next May.
It’s been incredibly rewarding to see people’s changing opinions of their local government
Hey all, I am a noob at building my own things. Recently, I have started building my own web app, which collects AI-related news worldwide. I also cluster them to identify trends within the AI economy. This was interesting to me as I do actively invest in stocks. The website also generates a newsletter, collects details related to new corporate deals announced, etc. It is collecting everything related to AI and the economy.
Now I feel lost, I don’t know where to go from here. I don’t even know if I am doing the right thing. What do you think? Is there any guidance or roast you can give? Here is the website https://www.racetoagi.org/
Here is the trends collection https://www.racetoagi.org/trends
Here is the deals graph https://www.racetoagi.org/deals
Finally, here is the newsletter https://www.racetoagi.org/research/newsletter
I'm working on a game for pocketstation (essentially Dreamcast VMU, but Playstation). It has the same cpu architecture as GBA but there are some unfortunate circumstances that requires me to modify LLVM for rust to use. Forces me to learn I guess
I'm looking to leverage the upcoming WebNN browser spec with my spreadsheet app. I think integration with vision, audio and language models opens a whole new world of possibilities as traditional spreadsheet apps lean more towards mainly numerical data.
Working on an MPC stack to make it easier for devs to integrate privacy into their stacks. As normal folks increasingly value the privacy of their data, developers will need to think about how they can build apps while guarding their users' data. We provide tooling for them to do this.
Still WIP but we are getting our first audit in the coming days!
Stoffel-Lang:https://github.com/Stoffel-Labs/Stoffel-Lang StoffelVM: https://github.com/Stoffel-Labs/StoffelVM MPC protocols: github.com/Stoffel-Labs/mpc-protocols Website: stoffelmpc.com
Added a fifth project this month. Most likely very unwise...
1. probe.bike - tell stories with your bike rides. It allows you to aggregate your cycling trip into one datapoint. Will likely break this out to skiing over the break and rebrand slightly. Adding yearly cards as we speak!
2. flopper.io - I'm seeing traffic rise and rise for this and it's been a great way to translate my every-increasing understanding of AI Infrastructure architecture to a new project. It acts as a benchmark website for GPUs and systems (e.g. Nvidia NVL72.
3. llmstxt.studio - still feel like llms.txt as an idea make sense - so hedged that and but let's see. Got my first customer this month. B2B and need more features/marketing.
4. rides.bike - the oldest - a catalogue or well researched cycling destinations and information about destinations. Will be adding more very soon!
Listing all the AI horror stories on https://whenaifail.com
I've really enjoyed writing blog posts recently. Not only is it a great way to flex your writing muscles, but writing about a topic, unsurprisingly, helps you understand that topic better too. I've had great conversations with friends about the posts I've written as well.
And sort of in that same vein, I've been developing my own static site generator that I eventually want to move my blog to. It's almost certainly going to be a worse SSG than every alternative, but it'll be mine and that's worth something in itself.
Plus it's just been fun to make! I wrote some gnarly code to generate infinitely nestable layouts that I'm kind of proud of. It's the kind of code that's really cool but you can only code on a project for yourself, because if someone else had to debug it, they might say some pretty unkind things about you.
Trying to make my Rust library `composable-indexes` more ergonomic. It is for indexing a collection on multiple dimensions in a type-safe and composable manner.
In other words, something safer & more concise than maintaining multiple HashMap's, but a lot less involved & simpler than an in-memory SQLite.
It's better explained by the example here: https://github.com/utdemir/composable-indexes/blob/3baa36762....
Feels like I'm working on a million things (between work, side contracts, and creative explorations). Recently a friend asked whether AI is helping or hurting my workflow.
And I realized I couldn't give a concrete answer. Lots of speculation, but I realized I didn't have hardly any real data. Inspired by Adam Grant's work on "rethinking", I'm _currently_ writing a tiny CLI to run self-experiments on my own productivity, auto-checking in / observing commits/code changes.
Goal at the end is to be able to test myself across different dimensions with "no AI", "moderate AI" (e.g. searching, inline assist), and "full AI" (agents, etc). https://github.com/wellwright-labs/pulse
https://fooqux.com/ - an experimental article aggregator about software development. For several years now, I've had a routine of collecting articles on topics that interest me throughout the week and then reading them over the weekend. To help organize and streamline this process, I created this website. The main idea is to gather tech articles in one place and process them with a LLM — categorize them, generate summaries, and try experimental features like annotations, questions, etc. I hope this service might be useful to others as well.
Building https://programmer.network/ with a hope to eventually gather like minded nerds that value their privacy, time, and who don't want to be victims of engagement algorithms. Building it live on Twitch past couple of years, with very limited amount of time, but it's fun, engaging and still enjoyable.
A prompt engineering tool that takes vague prompts and transforms them into context-rich JSON/XML structured prompts. Fully customizable and tracks prompting history safely in tab session, automatically injecting context and learns style.
Makes outputs of any AI so much better due to restructuring and breaking down requests into instructions AI can easily execute upon and mitigating risk of hallucinations. Perfect for complex tasks like coding and content creation.
Exists as a free chrome extension right now. Would love if you tried it and have any feedback!
Email me at [email protected]
https://github.com/cdlewis/snowboardkids2-decomp
A matching decompilation of snowboard kids 2 for the n64. Why this game? Well it's awesome but also I wanted to work on a decomp project from scratch. I've written several blog posts about my experience for those interested. I hope to do more in the future, probably with less of an AI focus.
* Using Coding Agents to Decompile Nintendo 64 Games https://blog.chrislewis.au/using-coding-agents-to-decompile-...
* The Unexpected Effectiveness of One-Shot Decompilation with Claude https://blog.chrislewis.au/the-unexpected-effectiveness-of-o...
I’m working on a deterministic execution layer for AI systems.
The idea is to treat LLMs as constrained components inside explicitly defined workflows: strict input/output schemas, validated DAGs, clear failure modes, and replayable execution. Most “AI unreliability” I’ve seen isn’t model-related — it comes from ambiguous structure and hidden assumptions around the model.
We’re exploring this through a project called FACET, focused on making AI behavior testable, debuggable, and reproducible in the same way we expect from other parts of a system.
Still early, but the goal is simple: less magic, more contracts.
https://ideawell.fly.dev/ - business idea generator based on the HN conversations (launched yesterday)
After reading an article about doing 10,000 pushups in a year (https://wjgilmore.com/articles/10000-pushups), I created "push10k", an iOS app to help me keep track and stay motivated. It's free (no money, no ads) in the iOS app store: https://apps.apple.com/us/app/push10k/id6754811078.
Petra (Obsidian CLI) and Petra-Bridge (Obsidian API for the CLI tool integration). https://github.com/H4ZM47/petra-obsidian-cli ; https://github.com/H4ZM47/petra-bridge
This was a "scratching my own itch" project that I cooked up because I was tired of Claude et al cluttering up all of my stuff with random markdown files. Just a simple Obsidian plugin to serve an API that the CLI tool can use to interact with the Obsidian vault. I use a Claude Skill to get the model to create all of those random markdown files in my Obsidian vault, and read from them when it needs context for something. It's working really well for me so far!
I am working on a text-editor that allows you to discuss specified file folders with AI agents for knowledge work. It is similar to vs code with chat but for non-technical users and not for code.
I feel like I am constantly fighting LLM interfaces to make available and organized the context needed for discussion. There just seems to be way to much copy pasta into and out of the infinite scroll interface. I also find the output tough to quickly edit and discuss with the chatbot.
It is simple enough but I couldn't find anything like it and it has quickly become one of my favorite tools. I am build over buy to a a fault, so if there is something out there like this already I would not be surprised.
Today is the start of Langjam Gamejam, a 7-day hackathon to build a programming language and then make a game using it. I'm ideating on what I'll build.
LLMatcher - blind testing arena to find which AI model actually works best for you.
You enter prompts, compare two anonymous responses, pick the better one. After voting, it reveals which models you preferred. Built it because model benchmarks don't match real-world preference, and blind pairwise comparison cuts through the hype.
I’m speed-running a bunch of new hobbies to teach myself how to make a physical game (basically its a ping pong paddle that tracks how often you hit a ball — like a “keepy uppy” game with scorekeeping):
- Arduino dev and circuitry
- 3D printing
- PCB design
- Woodworking
Its all a lot of fun and IMO a lot more approachable than it has been thanks to the assist from LLMs.
Working on promptfoo, an open-source (MIT) CLI and framework for eval-ing and red-teaming LLM apps. Think of it like pytest but for prompts - you define test cases, run evals against any model (OpenAI, Anthropic, local models, whatever), and catch regressions before they hit prod.
Currently building out support for multi-agent evals, better tracing, voice, and static code analysis for AI security use cases. So many fun sub-problems in this space - LLM testing is deceptively hard.
If you end up checking it out and pick up an issue, I'll happily send swag. We're also hiring if you want to work on this stuff full-time.
As a means to learn about both WebAssembly and Rust, I started writing a WebAssembly binary decoder (i.e. a parser for `.wasm` files) from scratch.
Recently it hit v2.0 spec conformance. 3.0 is next on the roadmap. (I'm executing it against the upstream spec test suite.)
I don't plan to make it a highly-performant decoder for use in production environments, but rather one that can be used for educational purposes, easy to read and/or debugging issues with modules. That's why I decided not to offer a streaming API, and why I'll be focusing on things like good errors, good code docs etc.
P.S. I'm new to the language so any feedback is more than welcome.
Cross-platform game framework for/in the Odin programming language. It's also the foundation for my first Steam release. The plan is to get something out on Steam, roll with the punches (bugs,) then open it up for general-use. I say "framework" instead of "engine" because the scope of the project is to make the decisions the beginners get stuck on and free them to make a game. That's a smaller goal than what you see with Unity, Godot and Unreal, but I am already at the point that I'd rather use my thing than Godot.
Basically personalized meal planning and grocery integration. Since the Show HN I posted a couple months back I've been incorporating user feedback to add things like meal prepping, better ingredient reuse across meals, and cooking style preferences.
One of the biggest points of feedback has been adding more grocery stores but I'm really limited by who has APIs to actually integrate with, which is basically just Kroger and Instacart. Walmart has an API but ignored my API access request. Would love to hear if anyone has ideas on how to approach this.
Pagecord!
An independent blogging and personal website builder. Source available (Ruby on Rails).
It’s not a novel idea but it’s gaining decent traction because it’s simple and (I think!) makes you want to write more. Which is basically why I built it.
Blog by email, custom domains, internal private analytics, theming and more!
Free forever plan, or only $29/yr for everything. Priced as I think personal/blogging sites should be. Everything is too expensive these days.
Trivyn: Ontology-first knowledge platform. Runs on a single machine, via a single executable. I wanted a simpler alternative to the large complicated enterprise products that tend to dominate this space.
I'm really trying to get a private beta out the door by Christmas. I do plan to have a free version for academic/personal use.
Backend is written in Rust, uses oxigraph for its triple store.
I'm working on a new kind of DAP (Digital Audio Player) with the focus being on a better visual experience to go alongside the music. Post going in-depth here: https://substack.com/home/post/p-181321780
I've been working on several internal tools that act like extensions for the django admin.
- https://github.com/yassi/dj-redis-panel - https://github.com/yassi/dj-cache-panel
This week I'm taking a break from my next project in this series (celery related) to try to participate in game jam related to programming language creation:
- https://itch.io/jam/langjamgamejam
I encourage others to participate I e
https://monofocus.lovable.app/
Yup, it's another task manager.
I made it for myself to help me focus on one task at a time, hence the name.
It implements my number one productivity hack of picking a task and setting a timer. Time spent on a task increments.
Data is stored locally in the browser although there is a sync option i wouldn't shake a stick at if I hadn't built this myself.
Plus it's a PWA! Those are lovely.
Plug-That-In [https://plugthat.in] (Mac App; Paid)
An annoying little laptop charging reminder utility that does the job.
---
There are times when I'm deeply immersed in focused work, a meeting, or engaging video content and end up missing the usual low-battery notifications on my MacBook.
When the laptop suddenly shuts down, it's followed by the familiar and frustrating walk to find a charger or power outlet. It can be annoying and occasionally embarrassing, especially when rejoining a session a few minutes later with, "Sorry, my battery died."
Over the past few weekends, I built Plug-That-In, an app that introduces "floating/moving notifications". These alerts follow the cursor, providing a stronger, harder-to-miss nudge regardless of what’s happening on screen.
The app also includes a few critical features:
- Reminder Mode: When the battery reaches critical levels, the app emits a configurable alert similar to a car's seatbelt warning, continuing until the battery is addressed.
- Do Not Disturb Settings: Customize alerts and sounds based on context, such as when system audio is playing, a video is active, or the camera is in use.
It grew out of a personal need, and I'm glad to see it used by over 50 people in the past month.
Working on https://canine.sh, an open source, self hosted PaaS for Kubernetes.
A big part of this was inspired by the last startup I worked at. In an effort to not deal with complexities of Kubernetes, we ended up on Heroku and was charged exorbitant amounts of money. One year spending close to 400k on Heroku alone, for what should’ve been 10-15k in cloud costs.
I think a big part of this is just making Kubernetes more friendly and easier to use for a small / midsized team of developers.
The goal is to make it easy enough for even a single developer to feel comfortable with, while also being powerful enough to be able to support a small team
TTRPG Narrative Engine — https://test.qualy.dev
I'm building a session prep tool for tabletop RPG game masters. The idea is to make a narrative engine rather than another static wiki. Most existing tools are great for storing lore, but they don't help you run the story. I wanted something that supports the "create now, refine later" workflow — get ideas into structure fast, then refine as you play.
Core features: - interconnected world-building (NPCs, factions, locations) and story-building (situations, fronts, clocks) - Bidirectional linking — connecting a story hook to an NPC makes that hook visible from the NPC's view - Clock system with milestone consequences that can spawn or edit entities - Situations fire different consequences based on outcome (players engaged vs. ignored the hook) - Material waste detection — flags under-connected content so you know what's prepped but unused.
The main workflow is mindmap-based. Each entity gets its own context layer showing direct relationships. (Soon available in demo version) Working on next: automatic player-facing content. As players complete situations, public notes from involved entities get published — so the GM doesn't have to maintain a separate campaign log.
Stack: TypeScript, Effect-TS, SolidJS, Cytoscape (graphs), Leaflet (maps)
The hosted version is rough — I've been using it to get early feedback from GM friends. Happy to hear thoughts from anyone who preps campaigns
I was thinking about FizzBuzz and thought it might be cool to benchmark various LLMs to see the highest number they could go before they got it wrong. FizzBuzz is cool because you can test whether the model's can generalize to any other game (divisors of 3 and 7 instead of 3 and 5 for example).
Fun, short and sweet experiment to run over the weekend, with some mildly interesting results :)
I'm working on an experimental display server, for educational purposes and fun:
https://terminal.pink/bgce/index.html
Or https://github.com/blmayer/bgce
The idea is to have the minimum needed for a usable graphical experience. So drawing to drm buffer and handling inputs basically. It's been fun to do.
I am build a toolkit for it too:
https://terminal.pink/bgtk/index.html
Or https://github.com/blmayer/bgtk
I think it is nice that we can just write to a buffer and it appears on the screen. Very little abstraction is needed. Hope you like it.
I also made some progress on my hardware projects, but I'll keep a low profile for now.
It's a travel tool for business travelers that figures out your suggested departure times for your entire itinerary based on predicted traffic patterns. Think Flighty but for all the non-flight parts of your trip.
You first build a travel itinerary with your legs - flights, activities, hotels (and hotel returns) and it tells you things like "leave your hotel at 7:40am" before your 8:30 meeting - in a single itinerary, no need to do the google maps acrobatics for every two items in your itinerary. While it's aimed at frequent business travellers I personally use it for all family leisure travel and daily itineraries around town as well - "do I have time for lunch at home after my son's class or should we bring packed lunch". I built it as during my time working in developer relations I traveled a lot, and always built unnecessary buffers and kept nervously glancing at my watch or phone to see if my planned time to leave still holds.
Tech-wise, currently it's Remix web app with a NodeJS/Fastify backend and Supabase for storage, and relying on google maps for route duration calculations. I want to expand it to native mobile clients in the future as well.
I am using it as playground on product thinking, ruthless prioritisation based on user benefit, figuring out unit pricing and economics, sensible architectural design, and exploring how including AI-enhanced features here and there can help make the product better, not just include them for their own sake.
Making music with the machine!
Album here: https://open.spotify.com/album/3e6k9eiGUlOBcoI2yd3DrM
Written about the process here: https://songxytr.substack.com/p/on-making-music-with-the-mac...
i’m working on Canva like app for automated image/pdf/video via API, also connects with n8n, zapier, make, airtable, pipedream etc.
it’s https://orshot.com
it’s being used by agencies and teams to automate pdf invoice/reports, instagram/tiktok/pinterest posts etc.
basically design a template, autofill the layers with your data from anywhere and generate visual content for marketing at scale
Currently working on Skyscraper, which is an iOS native Bluesky client: https://testflight.apple.com/join/RRvk14ks
Primary goal with the project was to create a Bluesky client that I wanted to to use. While the standard Bluesky app is fine, I wanted something more reminiscent of Tweetie, Tweetbot, Twitteriffic, etc. Something that feels at home on the iPhone. With the transition to Liquid Glass, felt like a good time to practice and get to experience the new UI with a new project.
Still in what I call "alpha", but pretty feature rich. Support push notifications, lots of discovery/feed following features, search tools, moderation setting management, post translation, and much more.
If you use Bluesky and are an iOS user, there's still space on my TestFlight and would appreciate any feedback or comments!
Working towards a handheld computer with a physical keyboard. Lots of examples out there (Hackberry Pi, Beepy, etc) but wanted to try my hand at it.
Along the way I found most of these use salvaged BlackBerry keyboards which are only going to become harder to find, so also on a bit of a side quest to build a thumb-sized keyboard from scratch. Got me into laying out and prototyping my first PCBs and learning about how these things are made - lots of fun so far!
Something cool I learned from tearing apart a BB keyboard: the satisfying “click” is just a tiny metal dome that pops and completes the circuit when pressed. Not news to anyone familiar with electronics manufacturing, but it was a cool thing to “discover.”
I'm working on adding an API and WhatsApp integration to my scam / AI detection tool - https://legitornot.co.za
https://github.com/sangeet01/limitnumen
As aspiring young man learning ai, I have successfully solved the theoretical limitation of embedding using hashing. Now working on turning that to RAG system as have solved the retrieval and now wanna complete the system.
Cheers :)
https://github.com/arcuru/eidetica
Eidetica - a decentralized database built in Rust, intended for local-first apps. It's still unstable but I'm progressing relatively rapidly. In the past ~month I have:
- Flown to SF to attend a conference in this niche: https://syncconf.dev/
- Added password based, transparent, end-to-end encryption
- Improved my custom CRDTs
- Added an index to store configs and metadata
- Built support for using sqlite + postgres for Eideticas backend (not pushed yet)
Once I finish the backend work I'll hopefully take a bit of a break though. I'm supposed to be retired.
https://bookpace.pages.dev
It's essentially a book progress tracker. There are many apps that allow you to add the books which you are reading currently, but not at what pace. It's simple, no complicated stuff, no AI shenanigans.
Created as I was overwhelmed by the number of books I want to read and thought it would be helpful to plan ahead.
You add a book name, number of pages and how many pages you want to read in a day. It calculates and gives you the number of days and on which date you will finish. It's also flexible to increase the number of pages so that it can recalculate.
It's a PWA for now. Still working on notifications and stuff.