• 🚀 Jev System-1 Model: Jev is a new type of AI model built for fast/inexpensive decisions. One or two orders-of-magnitude faster/cheaper than an LLM while still having a deep world model based on internet-scale training. There's some significant constraints and tradeoffs. Exploring those now
• ⚙️ Convex: My (previous) new favorite thing. A backend as a service (DB, Workers, Auth, File Storage, Scheduling) based on Typescript. Just write code and the schema/code/etc. just transparently shows up in the cloud. Perfect for AI side-projects as it all just works instantly. Everything is Typescript-defined, so you get excellent code-time error checking which is nice for agentic dev.
• Rust/tokio: I decided on Go rather than Rust 5 years ago when looking at modern System languages. I have no idea why, but it was a mistake. Rust is amazing and I love the power of Traits as a first-class language feature (rather than #ifdefs/macros/etc.). Learning programming languages isn't the value it used to be, but I want to get deep enough to guide my agents. And the interesting things I want to do in Rust (including making commits to Convex) use the tokio async threading library. These are both so huge I won't ever get super deep with them...

• 🌵 Prickly:Experimenting with Cactus AI speech-to-text and on-device AI inference
• ☑️ Semantically-Aware UI Toolkit: An exploration into using Jev to add intelligence to UI, both at the component and page levels. When a model is so fast and cheap, we can start considering including it in the browser event-interaction loop.
• 📹 Thelma: AI-powered video editing. I send it long rambling videos and it will cut into sensibly-edited videos at multiple durations. Captions, overlay graphics, story editing, fullscreen graphics. Can use hand-gestures as a meta-key to give voice commands...kinda. Built on top of FFMPEG and Remotion for Typescript-based graphics. It's all command-line driven and built for my workflow so this is definitely not going to be plug-and-play for anyone else
• 👾 konami: Jev-powered video game player. Uses Typesafe.ai's new Jev model to generate keypresses based on game state. Private repo as it contains code I do not own, but I'll split out the core service as an example someday
• 🏭 Forge: Automated AI content generation for the various apps below now automating technical SEO tasks for the website
• 🧭 ismitchmcconnella.live Squeezing 1,000rps out of a cheap VPS and Convex free tier? Public repo here mitch-dot-live
• 🧆 @lindale/porch Manages shared edge infrastructure to serve many domains/projects from a single cheap VPS. Everything here uses it and a $8 Digital Ocean droplet. MD Instructions provided so your agent can set it up in your repo.
• 💻 Convex Modifying the Convex self-hosted backend to add some debugging features and learn more about V8 and Rust
• 🐶 cafezoe.lol AI generated posts for your pets. Parts in the public lbbb repo here.
• 📖 nitenite.mom A daily bedtime story for your toddler with sticker-book animations
• 🤫 (and a saucy version for mom yet-to-be-named with audio narration)
• 🌮 Lindale Digital private repo work for clients
• 🤖 All the above built on an AI content generation lib I'm building with some novel features for storytelling (persistent characters, entities, de-slopification, etc.)
• 🕯️ kenzieskandles.com My grand-niece's side-hustle
• 🥧 squat-cobbler Collection of extensions and skills for the Pi coding agent that I will eventually use more. I really like Pi in theory but am really happy with the value, performance and output of Cursor's Composer 2.5 so I rarely use Pi. I expect with some upcoming persistent cloud agents I'm working on that most or all of them will end up being pi -p in some form

So many projects to start and never finish...