alt Hacker NewsAsk HN: What are you working on? (February 2026)
What are you working on? Any new ideas that you're thinking about?
Comments
I built a CLI tool in rust to find and remove duplicate files.
At work I sometimes download the same file multiple times to my machine and have to go through my downloads folder and manually remove the duplicates, so this is something I built to speed up the process of doing this.
https://github.com/jconvery1/hydra
I've found it quite useful.
I’ve been experimenting with a live win probability predictor for the 10-player arcade game Killer Queen. The goal is to predict the winner in a causal, event-by-event fashion.
Right now I’m struggling to beat a baseline LightGBM model trained on hand-engineered expert features. My attempts at using a win probability head on top of nanoGPT, treating events as tokens, have been significantly worse. I am seeing about 65% accuracy compared to the LightGBM’s 70%. That 5% gap is huge given how stochastic the early game is, and the Transformer is easily 4 OOM more expensive to train.
To bridge the gap, I’m moving to a hybrid approach. I’m feeding those expert features back in as additional tokens or auxiliary loss heads, and I am using the LightGBM model as a teacher for knowledge distillation to provide smoother gradients.
The main priority here is personalized post-game feedback. By tracking sharp swings in win probability, or $\Delta WP$, you can automatically generate high or low-light reels right after a match. It helps players see the exact moment a play was either effective or catastrophic.
There is also a clear application for automated content creation. You can use $\Delta WP$ as a heuristic to identify the actual turning points of a match for YouTube summaries without needing to manually scrub through hours of Twitch footage.
In my spare time https://bookclub.cloud.
It’s a tool that leverages your drm free ebooks to help manage your book clubs. The epub file offers more rich info such as:
- word counts per reading
- chapter selection
- the ability to highlight sections and share
- ai summarization and spoiler free discussion about the contents of a given reading
I use it for my own book club right now. I doubt I’ll be able to monetize the app due to the need for drm free epubs which is a pretty high barrier for entry to most non-technical users.
My long term plans would be to have an agent help readers learn hand in hand while reading. I’d like to have the agent facilitate deeper analysis by prompting the users and clubs with questions that encourage more critical analysis of each section. I’ve been building all the infrastructure for running the club so that’s the next more interesting step I haven’t explored yet.
Can't really share any useful link at the moment, but I’ve been working on a solution based on the C1001 mmWave sensor. It collects telemetry throughout the night, including respiratory rate, heart rate (BPM), number of turnovers, apnea detection, and overall sleep quality - you just need to hang it around the bed. The data is sent to a server, which generates a nightly report and passes it to an LLM for analysis.
I've decided to do it simply because of my grandmother: she dislikes bracelets and smartwatches because they’re uncomfortable and she often forgets to wear them. A contactless device could be much more practical for her and for many people like her.
Bitwit: Learn CS, logic, and math theory (_not_ DSA) with spaced repetition: https://bit-wit.com/
I was hit by a big wave of depression last year. That crisis gave me a couple of periods of hyperfocus, and I thought I might as well use them to try to improve my mind and my ability to reason.
I haven't had good results learning certain subjects, like CS/math, with Anki or other flash card systems. The only thing that ever seemed to work for me was doing a lot of problems with pencil and paper. But without problems exercising those tools or techniques, they tended to just evaporate from my mind.
So my idea was to combine spaced repetition and problem presentation, and create a system for generating problems and validating their solutions randomly and parametrically. So successive presentations of the same card would present a different problem... and the overall effect would hopefully be that you learn the invariants that each card is trying to present, instead of memorizing some combination of variables and values. Sort of MathAcademy but for CS.
I _think_ it's feature-complete at this point, though I'm still working on validating and publishing cards (and there are a couple of things I wanna do to improve security). I'm pretty early in the dogfooding process, and I'm still the only person using it, and there are definitely bugs. But if anyone would like to try it out, I'd be delighted to hear your feedback (email in profile, also on site). Just maybe go a little easy on me right now, lol :/
The Boolean Logic and Lambda Calculus subjects are free forever, and the first topic of each other subject is also free, so if you're intrigued you can use it for quite some time (literally months) before even needing to provide an email address. No need to pay for anything.
I've been building VVM: a language for agentic programs where the LLM is the runtime.
Most stacks today keep orchestration in Python/TS and treat the model like an API. VVM inverts this. You write a `.vvm` program, paste it into an agentic session with the spec, and the model simulates the VM and executes your program. This is configured via a plugin, no SDK or build step required.
```
agent researcher(model="sonnet", skills=["web-search"])
agent writer(model="opus")
papers = @researcher `Find recent papers on {topic}.`(topic)
# semantic predicate — semantic if condition
if ?`sufficient coverage`(papers):
report = @writer `Synthesize findings.`(papers)
papers = @researcher `Expand search.`(papers)
```
It's open source and current runs as a Claude code plugin and the repo includes 37 examples from basic agent calls to advanced patterns that showcase building resilient workflows.
Still pre-alpha -- the spec is fairly complete, but tooling is early. Mainly looking for feedback from people building agentic systems who've felt the impedance mismatch between imperative code and what agents actually need.
I’ve been building out https://hnarcade.com for the past weeks. Got a lot of good feedback from the ShowHN thread and others reaching out individually.
Since the ShowHN thread, I received more than 40 individual game submissions!
To give more exposure to some of the games launched during the week I also launched a newsletter. Feel free to check it out if you want to learn more about games shown over the week :)
Working on Memory Store: persistent, shared memory for all your AI agents.
The problem: if you use multiple AI tools (Claude, ChatGPT, Cursor, etc.), none of them know what the others know. You end up maintaining .md files, pasting context between chats, and re-explaining your project every time you start a new conversation. Power users spend more time briefing their agents than doing actual work.
Memory Store is an MCP server that ingests context from your workplace tools (Slack, email, calendar) and makes it available to any MCP-compatible agent. Make a decision in one tool, the others know. Project status changes, every agent is up to date.
We ran 35 in-depth user interviews and surveyed 90 people before writing a line of product code — 95% had already built workarounds for this problem (custom GPTs, claude.md templates, copy-paste workflows). The pain is real and people are already investing effort to solve it badly.
Early users are telling us things like one founder who tracked investor conversations through Memory Store and estimated talking to 4-5x more people because his agents could draft contextual replies without manual briefing. It helped close his round.
Live in beta now. Would love feedback from anyone who's felt this pain! :)
I'm one of the maintainers of PGlite - PostgreSQL in WASM.
We have lots of updates in the pipeline, just listing some of them:
- making it easier to upgrade PostgreSQL's version
- PostGIS - this is one of our most requested extensions and although we have it running server side (ie node, bun, deno) Chrome is holding us back from releasing it
- multi-connection multiplexing over a single instance
Native library is also on our radar, just needs more time...
I'm working on a (somewhat) realistic surfing game. Tired of arcade-style games, I decided to try my hand at something closer to the real sport, focusing on realistic breaking waves, speed generation and carving, rather than impossible air combos.
After one year of development, it's going better than I expected, so I'm considering building a demo to gather feedback and see if there's enough traction for working towards a Steam release.
Even if that's not the case though, it's been a blast learning about game dev in Unity/C#, as well as 3D modeling and animation in Blender!
A tiny web app for busy weeknight cooking.
You tap in what ingredients you’ve got, add a time limit + a couple of preferences, and it gives you 3 genuinely doable dinner ideas with step-by-step recipes (no “manage your pantry”, no endless scrolling).
It’s early, but people seem to like the “use up what you’ve got” angle. Feedback very welcome.
In my evenings and weekends I've been working on a photo sharing app for families who want to keep their photos off of social media but still share with family.
I'm expecting my first child soon so I am building it for me and my family first but if it solves a problem for me, maybe others will like it!
The basic idea is that you are uploading a curated set of photos you want to share, not your whole camera roll.
You can create one or several family groups that you can share individual photos or albums to. Members of those family groups can view, comment on, like, etc those photos.
You can also generate sharable links for people who don't have an account with a configurable expiry time.
It currently more or less works on the web but I am also working on iOS and Android apps since that is how my extended family would want to interact with it.
I'm not quite ready to launch it to the public but if anybody is interested in trying it out or offering feedback I can privately share it :)
https://azayrahmad.github.io/win98-web/ Another Web-OS remake of Windows 98 made with vanilla JavaScript. There are already a lot of Windows web remakes, especially in this age of AI. So for this passion project I intend to make it as accurate as it could ever be without having to emulate actual Windows 98.
Currently it has:
- Accurate recreation of Windows shell with start menu, taskbar, windowing system.
- Full desktop themes customization (color, cursor, sound, wallpaper, screensaver). All Win 98 default Plus! themes are included.
- Persistent local file system & mounting local folder as removable disk with ZenFS.
- Support playing Flash games and run DOS games (save game persisted). Yes, you can play Doom and copy your savegames to continue.
- Some accurate remakes of Windows 98 apps, some made by me (Solitaire games, Minesweeper, Notepad) some are existing ports (Pinball, JSPaint, Webamp, etc).
- Some other fun stuff
If you're interested in Windows 98, this is for you. You're also welcome to contribute or fork it to create your own version: https://github.com/azayrahmad/win98-web
I'm working on a language learning framework based on the ideas of comprehensible input and spaced repetition learning.
The idea is you take a book you want to read, and it gets translated but also rewritten to match your current learning level. And as you read/listen it introduces new words to learn, reinforced by spaced repetition.
We're taking a trip to France this summer and I'm hoping to have something usable for at least a couple months before we go.
Currently working on the mechanics of extracting content from ebooks.
I've been building several things lately in my spare time:
- After Actions (https://www.afteractions.net/) - The company I work for pays for retrospective software, and I figured I could make my own. So I did! Good for agile retrospective ceremonies
- Data Atlas (https://www.data-atlas.net/) - Right now, it's a basic JSON Schema form builder with a UI on top, but I intend to expand it and improve it to be more than that this year.
I've been working on a BuddhaBoard-like app and a few other projects, but nothing I'm prepared to share right now. Maybe next month :D
A django-based "offline" social media platform (i.e. forum) designed for access only from a single (isolated) WiFi network via captive portal techniques. The goal is a "digital bulletin board" that is geographically isolated and community-moderated (think Reddit's karma system but enough down-votes will delete your post).
Not anywhere near production-ready yet but if this is up your street and you want to contribute then I'd be grateful! I have the back-end stuff well under my belt but I'm allergic to javascript and my CSS skills are pretty limited also.
Also in the works down the line: a simple document management system (another Django app, probably) that allows basic but ISO-compliant document control (centralised access controls, automatic document-numbering, review/approval and draft/issue processes, configurable document indices, etc.) aimed at SME or contractor use. Everything else I've tried in this category of software is either abandoned (or otherwise stuck with a tech-stack from nineties), active but a PITA to ensure ISO-compliance, or hugely over-complicated (i.e. ideal for a big corpo but too much admin overhead for an SME) so I'm brewing my own that aims to meet the minimum feature-set needed for easy ISO-compliance.
Working on a little "passion project" that has ended up consuming a lot of weekends now, but it's been a lot of fun.
I've been building https://photoweather.app because I never end up having time to look at weather forecasts, which means I also don't go out with my camera enough since outdoor photography is quite a weather dependent activity.. so I'm trying to turn this around by having the app tell me when and where I could be photographing instead.
It's a bit of a challenge for sure, weather forecasts are not always the most reliable, not to mention learning enough about weather to forecast photographic opportunities.. but it's also been really enjoyable to finally build something real and something that I myself actually use all the time.
I am in the early stages on building a passion project called Metric Me - A dashboard for your body.
Over two decades ago I was diagnosed with high blood pressure (for which am I have been on meds for about 15 years). I also have low platelets (red blood cells, basically means that I bruise easily and that small cuts don't heal fast). At any rate, I do blood tests on a regular basis to keep things in check. I have been keeping track of test results, weight and blood pressure result for nearly 20 years, but the data lives in a text file on my desktop. I wanted to build something more substantial for this for quite some time now, so, this is it.
The second bubble there is a tool for 3D visualization and analytics of Claude Code sessions. The sample conversation is the one that made the tool itself!
That was a fun toy I learned a lot from. I’m not expanding that but am working intensely on the first bubble:
thinkt a CLI/TUI/Webapp for exploring your LLM conversations. Makes it easy to see all your local projects, view them, and export them. It has an embedded OpenAPI server and MCP server.
So you can open Kimi and say “use thinkt mcp to look at my last Claude session in this project, look at the thinking at the end and report on the issues we were facing”.
I added Claude Teams support by launching a Team and having that team look at its own traces and the changing ~/.Claude folder. Similar for Gemini CLI and Copilot (which still need work).
Doing it in the open. Only 2 weeks old - usable, but early. I’m only posting as it’s what I’m working on. Still working on polish and deeper review (it is vibe-crafted). There’s ergonomic issues with ports and DuckDB. Coming up next is VSCode extension and an exporter/collecter for remote agents.
Lately I’ve been interested in how budgeting apps model cash flow over time. A lot of tools are either expense trackers or monthly budgets that assume clean month boundaries, which breaks down quickly if you’re paid biweekly or irregularly.
Envelope budgeting (the method) works by allocating money up front so future obligations are actually covered before spending happens. But the hard part is handling income timing and paycheck variability without overfunding the future.
Anyways, I’m currently adding a cash flow detection algorithm to Envelope (the budgeting app) https://envelopebudgeting.com that only allocates paychecks to obligations before the next paycheck unless future funding is strictly required. That approach has avoided a lot of timing edge cases I kept running into.
I’ve been iterating on Toolbit (https://toolbit.app), a local developer utility app I initially shared a while back.
It started as a small attempt to stop bouncing between dozens of web tools for things like JSON/JWT, base64, regex testing, cron expressions, and similar “glue work” tasks. Over time it’s grown into something I keep open all day.
Since the last time I mentioned it, I’ve added quite a bit: • ~40+ tools now (regex tester, PDF merge, image conversion, cron builder, etc.) • workspaces for grouping tools around a task • tool chaining instead of constant copy/paste • snippets, history, and recent items • automatic detection (paste data, it routes you to the right tool)
It’s an Electron app and runs fully local. No accounts, no tracking, no sending data out. The goal isn’t novelty, just reducing friction in everyday dev work.
I’m still smoothing rough edges and figuring out where this is most genuinely useful. Curious how others here think about scope creep vs. “daily driver” utility tools, and what’s worked or failed for you in that space.
Sponder - Filter and transform RSS feeds or Podcasts (https://sponder.app)
Different RSS clients provide different filtering options, and lots of them limit you to a few keywords and/or put them behind a $7-12/mo subscription. I'm building Sponder so you can curate what you see, and it just presents another RSS feed, so you can keep using your favorite client but fill in the feature gaps.
Right now it can merge and filter by string or regex, and next I'm building (because it's what I want) history replay and smarter podcast rerun detection. it's new and I'm very open to feedback and feature requests.
I've been having a lot of fun modding 7 Days to Die (zombie game) using Claude Code. I have a strong development background but I've never worked on games before. Without AI it would be completely inaccessible to me without months of catching up.
7DTD is built on Unity Engine and modding is mostly done with XML/XPath and C#. I have yet to install Unity. I have a CLI setup, including a script to disassemble relevant 7DTD assembly DLLs into C# and copy other info from XML files into a directory in my repo.
This "refs" directory has about 800k lines of C# code and 300k lines of XML. Claude can figure out how to add a feature or fix a bug with a few minutes of searching the refs. At first it took 5-10 attempts to get results that actually worked, but now it's often 1-3.
Here are the mods I've released since early Jan:
- SteelUI: fixes for it to work with the latest game version[1]
- Smarter-Tools: a mod I authored from scratch to add a few tweaks to how tools work[2]
- More-Gore-Continued: has serious performance issues I'm working on - making progress[3] (adult only - requires sign-in)
If there's any interest I'm considering writing up all the details in an article and making the mod repos public.
[1]: https://www.nexusmods.com/7daystodie/mods/9386
Building ExoAgent: a security layer for AI agents that enforces data flow policy, not just access control.
The problem: agents like OpenClaw can read your email and post to Slack. Nothing stops Email A's content from leaking to the wrong recipient, or PII from ending up in a Slack message. Current "security" is prompts saying "please don't leak data."
The fix: fine-grained data access (object-capabilities) + deterministic policy (information flow control). If an agent reads sensitive data, it structurally can't send it to an unauthorized sink. Policy as code, not suggestions.
Got a working IFC proof-of-concept last week. Now building a secure personal agent to dogfood it.
What integrations would you want if privacy/security wasn't a blocker? What's the agent use case you wish you could trust?
Making improvements to https://engineering.fyi/ - an aggregator for big tech blogs
Added summaries so readers can quickly see if it's worth reading the entire article e.g. https://engineering.fyi/article/towards-self-driving-codebas...
ComputerPoker.ai
We trained PyTorch models on solved poker scenarios for post-flop, turn, and river situations. The planned "killer feature" is to give users feedback on their poker play in the flow of a simulated poker tournament or simulated cash game scenario. The goal is to play against "GTO Bots" (Game Theory Optimal Bots) to learn how to play closer to GTO.^1
Poker has been a passion of mine for a few years now, I find the game incredibly intellectually stimulating as well as a tremendous catalyst for personal growth, and this project has been a great way to channel that energy.
The web app uses Django/Channels/WebSockets. We've built an internal discounted CFR solver as well, hopefully building up to multiway scenarios in the future. The webapp is still in Beta/gated, and you're interested in learning more please email contact at surlesol dot com.
We are thinking of pricing $8/month or $74.99/year, with the rationale that this will be far less expensive than learning by experience at even micro stakes for online poker, with better feedback for learning, and at least we make it explicit that you're competing against bots ;-)
1. I am aware that GTO play is not always optimal, especially in live poker where live tells are available, and often exploitative strategies fare better than pure GTO. The target audience for ComputerPoker.ai is not hardcore poker pros, there's plenty of existing software for that, but rather those individuals looking to get acquainted with what GTO play "feels like." Then, with this knowledge in hand, knowing what the GTO play would be given various assumptions about our range and a reasonable opponent's range, we can deviate from the GTO play as deemed necessary.
- A phoenix/ecto inspired batteries included framework for Golang. Uses data-star for real time bindings (can do live view like things but my personal favorite is just real time form validation out of the box). Hot reload with templ, daisy, and tailwind (no npm required). Routes file provides metadata on routes so type safe route helpers are generated for views and handlers.
I always wanted a minimalistic CSS framework for my projects, so I started to create my own: THINK
THINK is a modern CSS-first UI framework built on semantic HTML, custom elements, and data attributes. Uses :has(), container queries, and density scaling. No classes, no build step.
It‘s work in progress but I‘m pretty happy with the outcome so far, especially the data table component and automated Insights. I know it‘s not AI driven - but it works pretty okay for quick insights on the loaded data.
Working on Whisperit v3: a document automation platform for lawyers, voice-first.
The idea: lawyers spend 60%+ of their time writing documents. We give them an AI assistant that actually understands legal context — case files, templates, emails, the full picture.
What we've been shipping lately on the v3 branch:
• Workflow engine — visual editor + execution pipeline for legal document workflows (think: intake → draft → review → send, all orchestrated)
• Case-context AI assistant — docked right into your case view, with file attachments, skill system, and next-action suggestions. It knows your case, not just your prompt. Think Pipedrive CRM for lawyer!
• Skills system — modular prompts lawyers can pick and inject (summarize deposition, draft motion, extract key dates, Claude Skills, etc.)
• Cloud drive sync — bi-directional sync with Google Drive and OneDrive. Lawyers keep their existing file setup, we plug in and keep everything in sync
• Outlook, Gmail, IMAP email integration — connect your inbox, pull relevant emails directly into cases
• Template multi-output generation — one case, multiple documents generated in one flow
• BYO AI provider — user/workspace/admin level settings for AI models. Some firms want Azure, some want Anthropic, we let them choose
• Canvas boards with AI context for visual case planning Stack: Next.js 16, FastAPI + LangGraph backend, Supabase.
We're a small team in Switzerland and across the globe, shipping daily. Target is Q1 for the v3 launch. If you work in legal or know lawyers drowning in document work: would love feedback or test our beta :)
A hobbyist analog computer I will open-source.
I have been through a number of iterations of adders, integrators… It's starting to coalesce now into a finished PCB. (Iterating is slow though at the PCB stage since we're round-tripping via Asian PCB fabs.)
As often happens, I want to learn about something (analog computing in this case) and I find that a deep dive where a kit comes out at the end is an enjoyable way to explore (I also get more experience with KiCad and, for this project, SMT).
The seed of this were a few articles on building a "lunar lander" circuit in early '70's electronic hobbyist magazines I found online. Pre-home-computer, how does one create a lunar landing simulation? Why with op-amps and panel meters of course.
And so that is where I began. But I'm way past that now though by making instead a general purpose analog computer.
Still working on:
A hobby project I started putting together late last year; a little spot on the internet for prayer and reflection.
https://dugnad.stavanger-digital.no/
A pro bono tech consultancy for local non profits. The idea is to help them use tech to better deliver on their mission.
Just today I finished off building a little PWA to help a couple of non-profits offload the admin of volunteer scheduling (mostly done through whatsapp, messenger, etc).
I'd recommend others try the same pro bono consulting in their local area, it's quite rewarding!
I'm still working on my Web Server Library .NET Core, the rewrite is almost finished (beta version)
I've been working on a browser extension to make Hacker News easier to use. No, not change the UX, but just some nice conveniences. Keyboard navigation, inline replies, dark mode, a nicer topcolors page, and many more features. I am hoping to add some social features, like being able to follow someone. All in a well-tested and extensible codebase that has minimal impact on the site. Open source, GPL...
Orange Juice
I've just started a new personal project, a C++20 library for running composable visitors over data documents and data models with JSON/CBOR semantics, DOM-less.
Basically, if you define a data model with bindings, you can inject data into it or extract data from it by running SAX-style visitors. You can use serializers/deserializers for standard formats like JSON/BSON/CBOR/CSV, or you can define custom formats for formating structured data however you want to. You can also run a serializer visitor on a deserializer to convert between formats. You can compose filter visitors to extract a subtree or filter out keys. And it's designed to fit on microcontrollers with very limited dynamic memory allocations, because it either streams data on-the-fly or works directly with the underlying data format in a big preallocated buffer.
I worked with libraries that offered a subset of these features before in my professional career (even built one myself), but recently I've had an epiphany (a document can also be used as a data model) that makes me think I can create something elegant and unique.
I’ve been working on https://og.plus, a service that creates unique Open Graph images per page on a website.
It does this by taking a screenshot of the page, but before it does that, you can modify what’s displayed in the screenshot with CSS, tailwind classes, meta tags, or HTML templates.
If you connect your website to it, the only thing you need to deploy to your web app are a few meta tags. The OG+ servers do the heavy lifting of processing the meta tags to setup the page, take a screenshot of it, and serve it up to the consumer.
The other cool thing it does is generate a different Open Graph images per social network so they all get an image for the exact size they works best in their previews. The CSS or HTML templates are aware of this too so you can display different content to specific social networks.
back working on my lighting desk, after a couple of years of hating it because the communications bus between the many different modules was flakey and so the whole thing wasn't fun to use. I bit the bullet last year and re-implemented everything with CAN-bus communications and it's actually fun to use now.
Current work has been improving boot time. Was nearly two minutes because of one board, and that's a long time for the lights to be out if you have to reboot during a show. I'd wanted to use buildroot to get a custom kernel that should boot much more quickly, but the buildroot learning curve was steep for me, particularly as I've no expectation of ever needing the knowledge again.
Independently but concurrently I decided I really ought to understand what all this AI stuff was about, for fear of getting left behind. That coincided with the release of opus 4.5, and holy heck has it made a difference! With a little guidance from me Claude got the buildroot environment working and the boot time down to less than 10 seconds. I've been _really_ impressed. I've had Claude write a few boring utilities that I could easily have done but Claude managed much faster and with less boredom on my part. Fortunately for my AI revolution I think I'm a better Business Analyst/writer than I am a coder, so it fits with my temperament.
https://github.com/tannn/freethoughts
Inspired by a TED talk I saw [0] where the researcher from Microsoft displayed a program with AI assisting with thinking while someone was reading and annotating a document. They claimed it was a way to sharpen critical thinking instead of killing it. They didn't release the product, but I figured it was cool and useful, so I've spent the weekend creating it. It's been a great way for me to practice using agents, and I've learned a lot from this process.
https://github.com/debba/tabularis
I’m working on tabularis, a lightweight desktop database management tool designed for developers. It provides a modern interface for managing MySQL/MariaDB, PostgreSQL, and SQLite databases through a native desktop application. Built using Tauri v2 (Rust backend) and React 19 (TypeScript frontend), it offers native performance while maintaining the flexibility of web technologies.
Working on the Mecha Comet (https://mecha.so/comet), a modular handheld Linux computer.
Biggest challenges: - How to explain the different use-cases/possibilities in a clear way - DX for any hacker who comes across the device with/without hardware experience
Recently, I got banned from Reddit for sharing my local news summarization website (www.cafelutza.ro) - for the Romanian market. So I figured you know what, I've been trying to bring this product to Reddit in the hopes of having better discourse around the news, but instead I realized, I was looking for smart discourse around a subject, which I haven't been able to find on Reddit or elsewhere, so I created Exppit (https://www.exppit.com) that basically gets experts to debate your topic of choice.
I'll admit it's terrifying to share this here because I don't know how to keep costs under control. For now only myself and my friends have used it.
Also trying to make a podcast out of it, which I enjoy listening to while I do some road trips: https://open.spotify.com/show/1fFwWMWJBJYIZmyz9cnrKB
I am working on an impression style city builder called Tutankhamun: Builders of the Eternal. I am the solo developer.
https://store.steampowered.com/app/4009620/Tutankhamun__Buil...
https://radius.to/ - a Meetup.com alternative of sorts - with fairer organiser pricing for smaller groups. I posted a Show HN [1] here a while back, got tons of great feedback, and have been slowly improving it since, with little marketing. Planning a re-launch here soon.
Developing LogiModel AI (https://www.logimodel.com/) which is an agentic supply chain optimization engine that forecasts demand, optimizes operations, and simulates scenarios to reduce costs while keeping your network reliable and customers satisfied. It integrates seamlessly with your ERP and document repositories to learn your business context, then acts as an intelligent decision engine, freeing you to focus on strategy while it handles execution.
Always interested in possibilities of LLMs interfacing with MIP solvers.
I’m building an AI emotional regulation coach for autism
Traditional therapy assumes you can identify what you’re feeling, which is exactly what many autistic people struggle with. Anna, the emotional regulation coach, addresses this dilemma.
I’ve used her daily for nearly a year and haven’t had a major dysregulation event since. Now I’m working on developing Anna further. The following post includes a detailed framework for building your own.
https://bettersoftware.uk/2026/01/24/how-to-build-ai-regulat...
A pricing API for the Counter-Strike 2 market that provides real-time & historical data.
The market is distributed across a bunch of 3rd-party marketplaces, and there's no 'simple' API that provides genuinely high-quality data. It's been a super fun project, and I've been able to learn about collecting & managing a high (to me) scale of data, building an API from the ground-up, and creating my first 'commercial' website.
Website is @ https://cs2.sh/
The API is built w/ Go & Clickhouse, which I've been super impressed with so far.
Web design is partially inspired by turbopuffer's site.
I have been working on a web app called Beval - Simple evaluations for your AI product.
In my day to day as a Product Manager working in a team that ships AI products, I often found myself wanting to do 'quick and dirty' LLM based evaluation on conversation transcripts and traces. I found myself blocked by 'Gemini in Google Sheets', it was too slow and cumbersome, and it didn't handle eval changes well. And because I was exploring, it wasn't helpful to try and set up something more robust with the team.
To fix the problem I eventually learned to call the OpenAI API in python, but I really felt that I wanted a 'product' to help me and potentially help others.
So this weekend I built https://beval.space
https://donethat.ai/profile/christoph
An AI based time tracker: reconstructs your day from whatever it sees you doing. Screenshot based but never stores them.
The same tech stack is pretty easily adaptable to openclaw tracking. If anybody would like to try, DM
Also looking into AI based security tools for monitoring security of DoneThat. Thinking of using zeropath would love to hear if people tried them / have other suggestions
Building PushMe: A service that sends you notifications for anything you care about.
I was wasting way to much time scrolling news. It consist of a crawling engine, de-duplication, llm creating a custom keyword filter and each event is checked against the original prompt for verification using a llm before sending it out.
Keywords suck, 381 of 388 potential matches end up like this:
"The event is about an OpenClaw Plugin Hub supply chain attack, not a data breach affecting hundreds of millions of users. The watch request specifies a massive data breach; the event does not clearly indicate that."
I am currently building an alethiometer from the phillip pullman book series: his dark materials
Specifically its a remake knowing what we know now having finished the two trilogies.
Instead of the three hands and the moving needle, its just one moving needle. Its made from wood and aluminium. The reason for aluminium is that in the 1920s (roughly the age that these books are set) is a wonder material, a bit like carbon fibre/titanium is now.
I have the working mechanism (controlled by 8mm steppers) I need to finish the case and dial, and paint it. I also need to shrink the controlling mechanism and design the voice->text->inference->output to 36 symbols logic.