Still on my AI extension for LibreOffice Writer. Growing ok, 1,700+ downloads since launch 3 weeks ago. Still no clue how to benefit from it, but I'm glad I've finally built something that's actually got traction https://extensions.libreoffice.org/en/extensions/show/99471
An experimental DSL for writing web applications. It's pipeline oriented, polyglot and loosely inspired by Bash pipelines. Rust runtime, fully-featured TypeScript language server (jump to def, hover, etc), full DAP debugging and a BDD-style testing framework built into the language.
GET /hello/:world
|> jq: `{ world: .params.world }`
|> handlebars: `<p>hello, {{world}}</p>`
describe "hello, world"
it "calls the route"
let world = "world"
when calling GET /hello/{{world}}
then status is 200
and selector `p` text equals "hello, {{world}}"
I saw the "TADA" post a week or so back (like "TODO" but you finished it—"Tada!") and I regretted not having done write ups for projects I did in 2025. So I'm doing that belatedly. (https://engineersneedart.com)
Also playing with building analog computers so I can understand them (and will not wait until 2027 to post about what I learned/did).
I used these two polarisers in my microscope and a magneto-optical sensor (which exploits the faraday effect), to visualise magnetic field lines of a magstripe card - https://www.youtube.com/watch?v=c8nM4Z-hkTw
I intend to try to use with a floppy disk, however it's not currently working for that.
An vscode extension to organize my lectures using a kanban style structure, saved as markdown files and allows specific md-formats to convert into presentations or other documents (currently marp (pdf, pptx, html) and i'm currently looking into pandoc). It's a vibe coded project i worked on in the evenings over the last 3 months.
What sets DeepFabric apart from other dataset generation tools is its ability to ensure high diversity yet domain-anchored relevance through unique topic graph generation algorithms. This guides sample creation to cover all necessary subtopics while avoiding redundancy, which is where other tools often fall short, resulting in model overfit.
Constrained decoding and response validation, along with real tool executions within isolated webassembly environments, ensure that generated samples strictly adhere to structured schema, variable constraints, and execution correctness, ensuring datasets have exact syntax and structure for use in model training pipelines. Tool definitions can be directly imported from MCP server schemas and then mocked, or rans as real life tool functions. Using real tools means the model has to adapt and correct when it makes the wrong choice or hallucinationates which makes for much better training data.
Once your dataset is generated, it can be automatically uploaded to Hugging Face and directly imported into popular training frameworks like TRL, Unsloth, and Axolotl.
Post-training, DeepFabric's built-in evaluation engine assesses model performance, whereby models prove their capabilities on unseen tasks derived from training splits—covering evaluation-only questions, answers, and tool traces.
I just launched the beta of my app that uses Elixir/Phoenix. Its a community driven aggregation of technology resources for Michigan, like companies in Michigan that hire tech talent, university programs, incubators in Michigan, etc. It also includes a newsletter thats already fully live and has over 10 subscribers. I'm gathering starting data and feedback now!
We’ve optimized the internet for producing information, not for humans consuming it. Most of us are overwhelmed not because content is bad, but because it’s all delivered in the same rigid format.
I’m working on unrav.io : a way to reshape any web content (article, video, or PDF) into the form that actually fits how you think (summaries, mindmaps, infographics, podcasts, chat, etc.).
We just launched a Chrome extension, so it’s one click on any page. No login, free to try.
I built an LED globe with a world viewing portal that uses a raspberry pi and pre-generated AI videos of places around the world. I'm working on upgrading it to do realtime generation of the videos and to improve the hardware so it is more durable and capable of being hauled around for community use.
I've been learning the basics of using Z3 by creating a solver for the daily puzzle game Clues By Sam (very fun game; https://cluesbysam.com/). Repo is https://github.com/DylanSp/clues-by-sam-solver. It uses Playwright to read clues and submit guesses; I've got it working for all 50 puzzles in puzzle pack 1.
I’ve been vibe-coding replacements for the tools I actually use every day. So far: a Notepad, a REST client, a snipping tool, a local gallery and lightweight notes for iPhone.
The motivation isn't novelty. It's control. I don't need ads, onboarding flows and popups, AI sidebars, bloated menus, unnecessary network calls , etc. A notepad should never touch the network. A REST client shouldn’t ship analytics or auto update itself mid-request.
No plugin system. No extensibility story. Just plain/simple software.
As I build these, I have been realizing how much cognitive overhead we’ve normalized in exchange for very little utility.
Our first offering is a tracker for makers, small businesses and contractors to show job status. Create a real time status page for your products, build trust and reduce customer inquiries.
We’re working on email notification support right now and have evidence (tracking numbers, job pictures, contracts/documents, etc) support coming next week!
I vibe coded https://domainhq.ai entirely in a few days with Opus, as in, this was a secondary focus and I never bothered to write any specs. Halfway through I had to ask it to convert it all to React because it was vanilla HTML and I didn't realize (that is not a flex).
I have a .ai domain that I feel is somewhat premium, and seemed to be getting lowball offers from domain brokers. And those I spoke to didn't seem like they were very up-to-date with market news and trends. Exploring services like Namebio were very limited and expensive; the subscription usage for building this was less than the price of a monthly membership (the domain is a different story!).
Domainhq.ai suggested its own domain to buy with its brainstorm feature, and tracks domain sales and expired domains, with reports generated by Claude as it perpetually analyzes price history over time.
The domain was a bit of an investment but I do like it, and the app itself is running 24-7 on a spare Macbook I wasn't using, so I plan to host it myself without paying for cloud servers, and the LLM features are using my personal subscription rather than API credits.
I do plan to add registrations and paid subscriptions (at a fair price, and will make things robust then); I get the irony here, but I also encourage anybody else who needs a niche tool to play around with their own vibe coding sessions to get something usable. Stuff like this does make me question the longevity of tech as a stable career.
Last month I worked on some new features (like unlocking units) and some fixes to my browser autobattler https://lfarroco.itch.io/mana-battle .
I've been working on adding an async PVP mode to it (using Supabase db and edge functions), should be up in the next few weeks.
Learned a lot about shipping Electron apps and using shaders with webgl, might write a blog post about it later
I'm building a training platform for cyber security and secure development practices. Mainly to address the junior to mid engineer gaps of "depth" across general cyber security and software engineering skills. I've encountered quite a few folk who have surface level knowledge but when having to problem solve production ready problems are blocked quite quickly, it's depth of knowledge which is missing. If anyone is interested reach out in a preview / providing some thoughts, reach out - my emails in my profile!
75 comments
[ 3.1 ms ] story [ 84.1 ms ] threadhttps://www.chromatose.app
For creating & performing unique visuals on the go.
Freemium w/ access to all features (just a 2 patch save limit).
And a game I made with my son, similar to Chips Challenge: https://reddot.adit.io
I plan to make this OSS, but currently offer a managed version.
Im excited to go down the idea maze and product iterations as I think a big bottleneck for useful agents in B2B is context.
Runtime: https://github.com/williamcotton/webpipe/tree/webpipe-2.0
LSP (with GIFs): https://github.com/williamcotton/webpipe-lsp/tree/webpipe-2....
Looks like:
https://github.com/ericfortis/mockaton/issues/2
Also playing with building analog computers so I can understand them (and will not wait until 2027 to post about what I learned/did).
EDIT: Maybe I'm thinking of the wrong person? Here's the link: https://www.ducktyped.org/p/one-year-of-keeping-a-tada-list
And the discussion: https://news.ycombinator.com/item?id=46354282
I used these two polarisers in my microscope and a magneto-optical sensor (which exploits the faraday effect), to visualise magnetic field lines of a magstripe card - https://www.youtube.com/watch?v=c8nM4Z-hkTw
I intend to try to use with a floppy disk, however it's not currently working for that.
https://github.com/ludos1978/ludos-vscode-markdown-kanban
Recently used the project to train a 4B model to outperform Claude Sonnet 4.5 and Gemini Pro 2.5 at Tool Calling. Colab here to run a free T4 GPU:
https://colab.research.google.com/drive/1EG1V40v5xkJKLf6Ra6W...
What sets DeepFabric apart from other dataset generation tools is its ability to ensure high diversity yet domain-anchored relevance through unique topic graph generation algorithms. This guides sample creation to cover all necessary subtopics while avoiding redundancy, which is where other tools often fall short, resulting in model overfit.
Constrained decoding and response validation, along with real tool executions within isolated webassembly environments, ensure that generated samples strictly adhere to structured schema, variable constraints, and execution correctness, ensuring datasets have exact syntax and structure for use in model training pipelines. Tool definitions can be directly imported from MCP server schemas and then mocked, or rans as real life tool functions. Using real tools means the model has to adapt and correct when it makes the wrong choice or hallucinationates which makes for much better training data.
Once your dataset is generated, it can be automatically uploaded to Hugging Face and directly imported into popular training frameworks like TRL, Unsloth, and Axolotl.
Post-training, DeepFabric's built-in evaluation engine assesses model performance, whereby models prove their capabilities on unseen tasks derived from training splits—covering evaluation-only questions, answers, and tool traces.
https://github.com/always-further/deepfabric
Upload a CSV or circle neighborhoods on Google Maps to build your address list (consumers or businesses). Printing and postage included in one price.
In the last 30 days I've added integrations for Zapier and Jobber (after adding Pipedrive, Zoho, and Follow Up Boss last month).
If anyone wants to help test these new integrations, I'll set you up on a special plan and let you send mail at my cost (roughly the price of a stamp)
There is no barcode as you bring your own glass jar and you fill it with pasta, rice, sugar ect.
The manager wastes too much time looking at what's inside and typing the code into his till
https://michigan-pulse.gigalixirapp.com
https://github.com/pajanowski/raymenuz
I want to use this for future raylib games as a developer menu.
I wanted to learn more Zig and be able to add or remove game state fields to and from the screen without having to recompile.
I’m working on unrav.io : a way to reshape any web content (article, video, or PDF) into the form that actually fits how you think (summaries, mindmaps, infographics, podcasts, chat, etc.).
We just launched a Chrome extension, so it’s one click on any page. No login, free to try.
https://unrav.io
The motivation isn't novelty. It's control. I don't need ads, onboarding flows and popups, AI sidebars, bloated menus, unnecessary network calls , etc. A notepad should never touch the network. A REST client shouldn’t ship analytics or auto update itself mid-request.
No plugin system. No extensibility story. Just plain/simple software.
As I build these, I have been realizing how much cognitive overhead we’ve normalized in exchange for very little utility.
https://jkl.io/
https://turboops.io/platform/public-tracker
Our first offering is a tracker for makers, small businesses and contractors to show job status. Create a real time status page for your products, build trust and reduce customer inquiries.
We’re working on email notification support right now and have evidence (tracking numbers, job pictures, contracts/documents, etc) support coming next week!
I have a .ai domain that I feel is somewhat premium, and seemed to be getting lowball offers from domain brokers. And those I spoke to didn't seem like they were very up-to-date with market news and trends. Exploring services like Namebio were very limited and expensive; the subscription usage for building this was less than the price of a monthly membership (the domain is a different story!).
Domainhq.ai suggested its own domain to buy with its brainstorm feature, and tracks domain sales and expired domains, with reports generated by Claude as it perpetually analyzes price history over time.
The domain was a bit of an investment but I do like it, and the app itself is running 24-7 on a spare Macbook I wasn't using, so I plan to host it myself without paying for cloud servers, and the LLM features are using my personal subscription rather than API credits.
I do plan to add registrations and paid subscriptions (at a fair price, and will make things robust then); I get the irony here, but I also encourage anybody else who needs a niche tool to play around with their own vibe coding sessions to get something usable. Stuff like this does make me question the longevity of tech as a stable career.