alt Hacker NewsAsk HN: What Are You Working On? (April 2026)
What are you working on? Any new ideas that you're thinking about?
Comments
https://www.github.com/neuvem/java2graph —- A semantic code graph for AI agents (for Java language)
The goal is to provide AI agents with deep understanding of the codebase and help them understand the context, not just text
Working on https://getAiNative.com
I kept seeing the same thing across engineering teams: everyone bought Copilot or Cursor seats, people used it for a few weeks, and then output didn't really change.
The tooling is good but teams are treating AI like autocomplete instead of integrating it into how they actually ship code.
My take is that AI starts making a real difference when you apply the same discipline you already have for software engineering.
Plan the work, implement, write tests, commit, open a PR, get it reviewed. If you just vibe code with no structure around it, you get messy diffs and hallucinated tests that nobody trusts.
So I'm building around that workflow. Think of it as giving AI the guardrails of your existing eng process.
Four months ago, someone at AWS asked me how Bedrock prevents LLMs from seeing customer data.Casual office conversation. Nothing serious. But it turned into a real discussion AWS's entire business is built on data protection, and here were all these giants calling AI agents the next big thing. Agents that, by default, see everything.I said AWS hasn't built that layer yet. But they will.That gap became astra.
Your agent gets tokens instead of real data. It reasons, decides, acts real values only resolve at execution. PHI, PCI, PII never touch the model context. Two lines of code, works with whatever you're already running.codeastra.dev That's what i am working on , did some test but need more customers review on that.
For the past 4 years I've been building a programming language reimagined specifically for games. It has automatic multiplayer, but also things like state, components, concurrent behaviours and reactive user interfaces baked into the language.
I have completed by SaaS after 3month. Now working on marketing strategy and directory for launch
Link is below if anyone interested. Catch The Signal: https://catchthesignal.com/
I've got a bunch of irons in the fire at the moment, most leveraging or built with agentic coding tools; my harness of choice these days is pi+codex.
- An internal apps platform built with bun, pg-boss, and railway
- A smart music setlist manager that downloads chord charts, creates spotify playlists, and automatically drafts emails with attachments and practice schedules
- A recruiting intelligence platform called Spotter that I built in a weekend[0]
- A voice-agent for a client in the banking sector, implementing deterministic workflows using openai realtime voice + finite state machines[1]
[0] https://www.youtube.com/watch?v=AOedMSddGDg
[1] https://blog.davemo.com/posts/2026-02-14-deterministic-core-...
I build a tool auditing CLIs to assess readiness for agentic use.
It's particularly focused on reducing token usage, self-discoverability, and flow safety.
a 90's Art Style (think Street Fighter 2) - Surf Forecast App.
I wanted a surf forecast app that i can look at glance, which "time-slot" of the week is good enough to go surf.
And I wanted it to look like nothing else out there, at least surf forecast wise
Building Tap — turns browser automation into deterministic programs. The idea: you forge a tap once (AI-assisted inspection of the page), then run it forever at zero AI cost. Working on the Chrome extension + MCP integration so Claude Code and Cursor can use it directly.
The interesting problem is reliability: sites change, taps break. So the second half of the project is doctor/heal — automatic diagnostics that cross-validate what the tap extracts against the page's structured data (JSON-LD, ARIA). Feels like writing a compiler that also has a test suite built in.
I built a MacOS-native app [1] to control Positive Grid Spark amps [2], without needing a phone.
Official app is mobile-only and clunky, and the workflow is awkward if you're sitting at a desk. Hardest part has been maintaining compatibility across amp models. Small protocol changes or optimizations I make for one amp can break another. That means I have to do a lot of manual testing before every release. So I'm trying to think of an emulation layer or test harness I can build to make my life easier. Happy to hear suggestions there.
About ~50 people are using it so far, and main feedback has been that it's much faster and more reliable than the official app.
[1] https://tonepilot.app [2] https://www.positivegrid.com/products/spark-2
I am building AdamEyes a horizontal infrastructure to mimic human like vision, over which any robotics or computer vision applications can be build. I believe that natural navigation and sense of enviornment comes from vision, so why expensive system is required in robotics and machine perception, for basic applications like Robotics Navigation, drone surveillance, elderly care CCTV camera, Intelligent CCTV camera. AdamEyes enable any camera to have intelligence and behave like true human vision system. Please join us on LinkedIn on "The AdamEyes Project". Visit Our LinkedIn page: https://shorturl.at/Tt2E4
I'm working on curyo.xyz. The main idea is to create a new rating system for the age of AI, since we think we need better/new quality signals for the web.
Current features: Using zero-knowledge proofs for a proof of personhood (no personal data shared), Reputation as a conviction signal and gamification (gain reputation), Blind voting phase to fight herding, Using a public ledger/blockchain in the background for transparency
We know that the current implementation is far from a perfect rating system, but that’s why we are here, trying to get as much feedback as possible and finding the right balance between hard to attack and user-friendly.
I’m working on Flowtelic. A workflow driven note-taking system that aims to get you thinking deeper, but also help you work on the most important thing next if you’re stuck. While not essential, it’ll be enhanced with a local first AI approach.
We're still building https://shoehorn.dev/: an Intelligent Developer Platform (think Backstage, but opinionated and simple). With Shoehorn, you just run the thing.
"The irony of Backstage is that it was created to prevent teams from having to reinvent the wheel every time, building and maintaining their own developer portal. But that's exactly what everyone does with Backstage."
We wanted something you configure,deploy,update. thats it.
service catalog, GitHub crawler, K8s entity discovery via k8s-push-agent, Forge + molds (scaffolding/workflows, like Backstage templates), governance, scorecards, cloud provider resources, license management, event based notifications, team-context aware, API keys with scope auth alongside session RBAC. CLI and Terraform provider too.
We're aiming to release Beta end of April.
I built a platform to learn how to build personal AI agents and test them with fast feedback. It is free for individuals and small teams.
Platform deterministically generates tasks, creates environments for them, observes AI agents and then scores them (not LLM as a judge).
We just ran a worldwide hackathon (800 engineers across 80 cities). Ended up creating more than 1 million runtimes (each task runs in its own environment) and crashing the platform halfway.
104 tasks from the challenge on building a personal and trustworthy AI agent are open now for everyone.
To get started faster you can use a simple SGR Next Step agent: https://github.com/bitgn/sample-agents
TypeQuicker (https://typequicker.com) - personalized typing app.
I believe anyone can learn to type fast - I think it just takes the right tools to make it interesting enough for the users to use consistently
Now ready to release https://mealplannr.io. The end game is no/low touch weekly meal plans sent directly to your inbox, with meals from the chefs you follow - with none of the hassle around planning the meals, shopping list etc (which I spend hours doing every week).
An important feature for me was improving the recipe discovery experience, you can build a cookbook from chefs you follow on socials (youtube for now), or import from any source (Web, or take pic of cookbook etc) - it then has tight / easy integration into recipe lists.
Utilising GenAI to auto extract recipes, manage conversions, merge/categorise shopping lists etc - as-well as the actual recommendations engine.
If anyone is interested in beta testing / wants to have a chat I'll look out for replies, or message [email protected]
Building Gordon (https://trygordon.ai/) because most companies “have security” but no idea what their actual risk is. We’re building a platform that helps you pass audits faster, lower cyber insurance premiums, and makes cyber less of a fire drill.
Its like one place to see risk, catch threats, test what breaks, track vendors, train people, and not get hacked at 2 AM.
I continue to add more features to my search-and-replace tool for Windows (https://www.abareplace.com/). I initially built in for myself and it was one of the first incremental grep implementations (allowing you to see the results as you are typing the search pattern). Now it supports various formats like Base64, URL encoding, or timestamp conversions. With one-liners, you can add width/height attributes to <img> tags or insert file contents.
Tinkering around building https://viberglass.io. It started off as a way to allow my wife to keep her website up to date without any coding skills, but ended up growing arms and legs.
It is a hosted ticketing system/agent harness platform with integrations towards other ticket systems and chat apps. It allows triggering agentic (coding) tasks without the need to context switch and/or know anything about installing the wanted tools, SDKs, IDEs etc. Ephemeral workloads in isolated containers or cloud compute. Trying to help commoditize small scale development tasks and and prevent them from getting lost in the void of the backlog.
Open source with local or AWS self-hosted, full IaC attached.
I am building localgcp - This is localstack / floci equivalent for Google Cloud platform : https://github.com/slokam-ai/localgcp - currently supports 14 services, including Vertex AI and BigQuery.
Part of building this, I decided to build a BigQuery emulator from scratch and learned a lot about GoogleSQL (previously ZetaSQL) along the way: https://github.com/slokam-ai/localbq
I plan to maintain and improve this going forward. The goal is to see how much can emulators actually do.
Website: https://localgcp.com/
Working on Kernel, a GSAT vocab study app for Taiwanese students. A lot of exam prep still means paying cram schools for structure and linear repetition. We’re trying to turn that structure into software that knows what you’re about to forget and what to review next.
app store: https://apps.apple.com/tw/app/kernel-%E8%83%8C%E5%96%AE%E5%A...
viral launch post that brought in ~1700 users in 2 days: https://www.threads.com/@sean_hsu_13/post/DW8nBzDjV8T?xmt=AQ...
I'm working on LookAway, a Mac app that reminds you to take breaks from the screen at the right moment instead of interrupting you at random. https://lookaway.com
Right now I'm focused on the stats side. It already shows how much time you spend in each app, and I'm adding website tracking too, which should make the picture much more useful.
I'm also working on better break timing for dictation. LookAway already delays a due break if you're in the middle of typing, so it does not interrupt at a bad time. Now I'm trying to extend that same behavior to dictation as well, which turns out to be a pretty interesting detection problem because it overlaps with some of the other context signals I already use.
Most of the challenge is making it smarter without making it feel more intrusive.
Still working on a specialized slicer for a new 3D metal printing process. Mostly addresses material safety, model accuracy, and bringing technology within hobby budgets.
Also, cleaning up a microscope 4-axis micro-positioning stage project control-loop.
Finding spare time to deal with a backlog of various other small projects. =3
Some of my recent projects in case anyone finds them useful and/or interesting:
Automatic differentiation library in Clojure (https://github.com/cloudkj/lambda-autodiff) - inspired by Karpathy's `micrograd` from a few years ago; dusted it off recently, fixed a few issues, and was able to use it to implement a version of `microgpt` - https://cloudkj.github.io/lambda-autodiff/doc/examples/gpt/
PG&E "Share My Data" self-access library (https://github.com/cloudkj/pgesmd_self_access) - been tinkering with various home automation and monitoring ideas, and was able to get an end-to-end prototype for ingesting and visualizing PG&E meter data using a combination of the (forked) aforementioned library, an old circa 2015 Raspberry Pi, and a handful of dollars spent on AWS services (certificate manager, load balancer) to get the full mTLS PG&E integration working. Probably deserves a blog post to document all the gory details.
Geo data mashups (https://github.com/cloudkj/snowpack) - small frontend utilities to overlay custom data on top of each other; was able to satisfy two recent personal use cases: (1) visualize snow depth across California ski destinations and (2) heat map of national park traffic by entrance. Previously posted at https://news.ycombinator.com/item?id=46649103
REST interface for Gymnasium reinforcement learning (fka OpenAI Gym) (https://github.com/cloudkj/gymnasium-http-api) - simple wrapper around the forked version of OpenAI Gym to allow for language-agnostic development of RL algorithms.
I built an AI that turns YouTube videos into interactive tutoring sessions
Paste a link → AI breaks it into sections → teaches you on a whiteboard with voice → quiz + flashcards at the end.
It's free to try while in beta: https://www.pandio.online
I own {{mylastname}}.com and have been making a small social network for everyone who shares my surname. (it's rare) I'm using Facebook's NLLB to translate posts to different languages since we are mainly split between the Americas and Europe.
Building Warp (https://warp.thegeeksquad.io), a storage engine where each entity is its own actor with its own SQLite shard.
I've been doing DDD and event sourcing for years but always had to squeeze aggregates and domain events into Postgres tables. I kept looking at what scaling would mean with CockroachDB or ScyllaDB and it scared me. So I asked what happens if you just make SQLite the storage and let the BEAM handle concurrency, one actor per entity.
Turns out it works pretty well. 1.5M events/sec on an M1 in Docker with 5 cores. ScyllaDB on the same hardware does 49K. Written in Gleam, but there's a TypeScript SDK if you just want to use it from Node.
I am working on Tuumik - self hosted time tracking and in/out board. The project initially targeted law firms specifically (I was an attorney for a decade before coming back to software development). But Tuumik is suitable for any team working remotely.
Demo in browser (no registration required, just jump straight in): https://demo.tuumik.com/start-demo
https://www.tuumik.com https://github.com/tuumiksystems/tuumik
If any HN reader wants to give it a spin, hit me up at support at tuumik.com.
I am formalizing as patents a series of developments in drone delivery which stand to massively alter the status quo with respect to retail deployments. This, and a portable demonstration, are targeting fundraising for go to market on our autonomous restaurants, which have approximately 10 years R&D. I'm looking for $100M with around $30M spoken for. On that main business we are objectively ahead of all of the early stage players in terms of technology on footprint, automation level, and per site capex, plus will have far greater scalability. In other words, we stand to have a real, venture-scale return while the early stage players don't. https://infinite-food.com/
Outside of the day job (PM at an enterprise SaaS company), I've been building an AI-native CLI for Todoist [1]. Started to solve a personal problem, automating action item extraction from my Obsidian notes, but it's become something bigger. The CLI treats both humans and AI agents as first-class users: TTY-aware output, a schema command for agent discovery, idempotent operations, that kind of thing.
It's been a great excuse to get back to my roots as an engineer and lean into some of the newnew with Claude Code. Learning a ton, having a blast, and also enabling being (marginally) more productive with my actual work day to day.
Data support tool for pharmacists to identify savings and best value opportunities for their local health system (NHS/UK)
I'm a pharmacist, worked in the community for 5-6 years before moving into medicines optimisation, which in short is focused on ensuring we use the right medicine at the right time to get the best return on investment in terms of £/patient outcome.
Been a hobby coder for about a decade now, but this is my first attempt at a full stack application (airflow, db, backend, frontend).
Mobile formatting is a little bleh and there's some obvious issues. But it's been rather nice setting up something a bit more rigid/resilient than my previous clandestine approaches
I am working on a Dungeons & Dragons combat tracker: "Top Of The Round".
For those DMs that use tools like these, my app sits between Shieldmaiden and Improved Initiative in terms of features/complexity. I tried to offer as many features as possible but "hide" them in a way that makes it easy to understand the most important information like initiative order, health and conditions, stat-blocks. But then I added many buttons with keyboard shortcuts and a quick-access command-palette (think MacOS Spotlight or Alfred on Linux) that lets you access even more commands and features just by typing.
It is in beta, free and and you can check it out at https://topoftheround.com
I'm working on an interesting infinite canvas calculator called https://minusplus.app/. This is an infinite canvas, where you click anywhere and type numbers, Press space or enter, and it will automatically calculate. You can tag #, add comments, and I have added some great keyboard shortcuts, which make it efficient to calculate.
I am an SAP Finance consultant, and I have to do a lot of calculations daily. And after using this, I can't go back to any other normal calculator. :-D
I'm working on https://suggestionboard.io, a live polling/feedback/Q&A webapp that doesn't require an account. Just launched the first version, now looking at the market and making small improvements.
A runtime for a long-lived LLM agent with ambient continuous self-perception, persistent memory, defined authority, domain-specific autonomy, and forensic accountability, all in a long ongoing relationship with a human. I call this type of system an Artificial Retainer, a non-human cross between a guide dog and someone like your accountant or lawyer. It is not designed to be your friend, but it could be a valuable colleague. Think of this as an attempt to build a trusted stable agent with a stable character that could last decades.
I have been building an agentic code IDE (like everybody else :)).
Native cross platform app coded in rust + tauri.
I prefer using it to the other agentic code apps I have used. It has multi tab worktree isolated agents, sandboxed tools, git integration, built in code editor (with inline generation), searchable document support (i.e. upload your docs, datasheet and you or the agent can use them) and even built in local image generation (using stable diffusion and flux schnell) and asset handling for game developers. Oh it also has a remote feature so you can share the gui or deploy it on a server and access it on the go.
Working on adding text to 3d also.
It is a hobby project that has grown quite large. Feel free to try it out.
I forked a simple DNLA server which has been dead for 10 years and updated it to .Net 9. The goal is that hard drives are too expensive now and so it's better to spin up media servers quickly and temporarily using whatever hardware you have on hand.
It currently builds and announces itself to my TV (can see the server in Roku Media Player) but crashes because the http server implementation is homemade and out of date. Copilot generated some options and I will be plugging an implementation of a sockets-based server in the next couple of days.
Building Atlas, a recovery analytics platform for athletes.
The core frustration: Apple Watch collects HRV, sleep stages, respiratory rate, blood oxygen, resting HR. Apple does basically nothing useful with any of it. You get ring animations and step counts.
Atlas pulls all of that together and turns it into two scores: recovery and training readiness. The point is to actually use the signal your sensors are already collecting and ensure when you train, it matters. It’s like Whoop, but actually works.
iOS app is live (finally!). Happy to talk shop.
I'm working on a version 2.0 of an app that's been out for a couple of years. I won't link it from here, because, unlike almost every other software company in the world, we are not interested in MOAR UZERZ. We provide a specific Service to a specific demographic, and they know how to find us, just fine.
This project brings in a lot of AI support. It's made a massive difference. The original project took two years to finish (actually four, but we did a "back to the ol' drawing board reset).
It looks like this may only take a couple more months. I've been working on it for two months, already, and have gotten a significant amount done. The things that will slow it down, will be the usual sand in the gears: team communication overhead. Could stretch things out, quite a bit.
Writing. I publish one long form essay a month with two published thus far. The third one is in editing stages. An enjoyable experience moving from internal notes to outward expression.
https://larm.dev, an uptime monitoring service with a focus on reliability and reduction in false positives. I’ve been building it for myself really but I figure it’s worth sharing it with people in case someone else finds it useful too.
It’s also a lot of fun to work on. Phoenix LiveView dashboard, go probes running on 4 continents, connected to the backend using websocket tunnels. Clickhouse for reporting. Even did a CLI and an MCP for fun.
You can take the probes for a spin with the free response time checking tool and see how fast your site is https://larm.dev/tools/response-time
Still working on my urban tree visualization! Spent some time polishing the ingest pipeline to make it easier to add new cities, added a genus/species level view to aggregate across cities, and added in some basic imagery so I can see what species are. Thinking about adding in a end-user facing ingest pipeline so I can add some trees I like that I see on my walks. Probably need a performance pass to since I'm scaling up the volume quite a bit.
https://greenmtnboy.github.io/sf_tree_reporting
Posted in last thread when it was SF only: https://news.ycombinator.com/item?id=47303111#47304199
I’m working on https://chess67.com, software for running over-the-board chess (clubs, coaches, tournaments).
I started this after volunteering at my kid’s tournaments and seeing how fragmented things are: • registrations in Google Forms • payments via Venmo/Zelle • pairings in SwissSys/WinTD • communication across email and text
Chess67 aims to unify that: • coaches can sell lessons and manage scheduling and payments • clubs can run events and communicate with players • tournaments can handle registrations, with pairing and USCF submission in progress
Still early. The main challenge is not building features but matching existing workflows, especially Swiss pairings, which are more nuanced than they look.
B2B SaaS to host 3D scans of DataCenters and industrial plants.
Basically a google streetview tour of your Datacenter or large industrial plant.
You can do some nice things like draw 3D linework to trace the paths of pipes, conduits, eg : https://youtu.be/t8nRhWUl-vA add notes with markdown and html links at useful places in the 3D space.
We have add-ons for generating an 'xray' view floorplan to make it nicer to navigate a large space.
I think we are the first to have a web uploader that can preview and import .e57 panoramas, directly in the web page [ and skip the points if you dont need them ]
Currently in use by a telco in the Americas.
Documenting all publicly accessible stained glass as possible. App is built with vanilla HTML/CSS/JS. Nothing fancy!
Hoping to release a beat tape. I've given up on trying to create new apps to try and get VC money. I tried this, often with exploitive co founders who expected me to basically make Facebook, but BETTER in a month for 3% of their company which doesn't exist.
I also make small games with Godot.
Create StarDict files from Wiktionary HTML snapshots for KOReader: https://xxyzz.github.io/wiktionary_stardict/
I'm surprised that no one has done this so I decided to give it a try.
I am building a dev envirnmont to use Claude Code (or other agentic coding tools) for electronics development. I give Claude access to my lab equipment like my scope. That way I can let claude programm an MCU and at the same time verify that it creates the right output. Also, it can correlate spice simulations with the actual measurements. Which is often tedious to do "manuall" and verify how good my spice simulations are. I am going to record a longer demo, but here is very short video I made about it: https://lucasgerads.com/blog/lecroy-mcp-tutorial/
I am building nao, an open-source analytics agent. With nao you can connect all your sources of context (from warehouse to your knowledge base) using a CLI, and then you have an agent that sits on top of it.
You can use the agent from any client (web, Slack, Teams but also other harnesses).
We think most of data analytics work will be transformed (and is already being transformed) from SQL monkeys to chat to analyses, but all the UI/UX are not designed for this and this is what nao is, being open-source because knowing how the context is managed is key.
With nao you can have a conversation and then shared the output on the form of a story, that can be either static or live replacing what old dashboards were.
We are close to 1k stars on Github: https://github.com/getnao/nao