NotifyButton - A simple script on the frontend of your site, a complete SaaS platform on the backend for DSA compliance.

If you operate in the EU and want to avoid heavy fines, this is for you. Once integrated, it allows users to report legal content issues directly to you, which you can then manage via a dedicated dashboard following official EU procedures. Without such a system, users are much more likely to file complaints through official state or EU channels, which can trigger investigations.

https://notifybutton.com/

Building ConvoLens [1] - app to explore the content of video interviews from YouTube channels I like (such as Dwarkesh Patel's): research by keyword, RAG, visualization of discussed topics with a 2D projection, semantic graph, and the possibility to generate a new video or audio from a playlist of video segments.

Supports only YouTube as the data source, and Gemini 3.1 Flash Lite for processing, but it can easily be tweaked. Runs locally with Docker compose.

[1] https://github.com/didmar/convolens

I'm doing crystal bindings for pipewire and wayland clients. It is nice to be able to do so much with so little. github / yxhuvud / [pipewire_cr | wayland_client]

I wanted a real native app (iOS/macOS) as a client for my agents and to be able to truly control / mange them from it. So, think Claude Code remote but not just Claude and a proper native app. Or the Codex app but actually native.

The server is a rust binary so you can toss it on any container/computer and connect to it in the app.

My philosophy isn't to replace my other tools I love like emacs, ghostty, etc. But I am taking a stab at "real time code review" and have some crummy magit-like code review built in that I need to revisit.

https://github.com/Robdel12/OrbitDock

ChatShell (https://github.com/chatshellapp/chatshell-desktop) — open-source desktop AI agent built with Tauri 2 + Rust. Ships with 9 built-in tools (web search, bash, file read/write, grep, etc.) so the AI can take real actions from the first conversation. No plugins, no config. Supports 40+ providers, MCP with OAuth, and a skills system. Apache 2.0.

Today working on adding chat history search (FTS5) and OpenRouter Nano Banana 2 support.

I am working on creating an Even Driven Architecture framework for Kotlin.

I went through the Software Architecture Patterns for Serverless Systems book, which I think it is fantastic. I learned a lot but I still had a lot of doubts to actually use the ideas in real life. So I started dissecting the companion framework, which is in written in Typescript. I have been going piece by piece and converting to Kotlin which I think it is more expressive (and fun) and it is allowing me to understand how everything fits together.

Typescript framework: https://github.com/jgilbert01/aws-lambda-stream

I started working on an "anti-habituation" motivational reminder app:

https://appreciate.srid.ca/

https://github.com/srid/Appreciate

(Note: Only the macOS version is tested extensively at the moment)

https://www.riotiq.com -> An AI model to predict human IQ with 90% accuracy. Our research team first needed to build the only ever professional online IQ test to collect the massive amount of raw data needed to train a model. We are starting model training now:

I’ve been working on an open source tool that turns your Kubernetes into a Heroku like PaaS — https://canine.sh — for about two years

A problem that we had at my last startup was that we got stuck between not wanting to spend too much time on devops, and getting price gouged by Heroku.

We were too big for the deploy to a VPS type options like coolify, but too small to justify hiring a full time Devops.

Eventually a few of us had to just suck it up and learn Kubernetes properly. Was pleasantly surprised how elegant it all was.

I was surprised there wasn’t something that “just worked” and plugged into our Kubernetes cluster, made it user friendly, teams, roles, etc.

When I discovered that some local llama.cpp can OCR PDF images generated by TeX, I started to revisit literate programming defined by Donald Knuth and explore using PDF as the source of truth artifact (instead of Markdown or program source code itself) for LLM to consume.

I only got to the point of having code and data as \verbatim in \LaTeX. Next step is CWEB.

Here is an example (with C and Rust code in \verbatim)

https://ontouchstart.github.io/rabbit-holes/llm_rabbit_hole_...

The ultimate goal is machine and human readable proofs on algorithms.