Hi David,

Fran here. I'm building a niche tool for designers and architects to create AI renderings in seconds with no prompting skills, no ComfyUI installs, no chasing the latest Nano-Banana, Flux, or any other GenAI image model. I called it RenderAI. Want to make it simple and fast.

We already have thousands of users, which confirms there is a real need. Still, I’d love to get honest feedback. How can we stand out? Feel free to be blunt.

Check out https://RenderAI.app if you’re curious.

Just published the Halo vision headphones [1]. Currently working on reducing parts where possible, tuning audio quality and video stabilization.

https://haloheadphones.com

Working on AI interpretability infrastructure! Starting with hallucination/failure mode detection and causal tracing in transformer-based foundation models. It's non-SAE, domain-agnostic, and works for any kind of tokens or sequence data (e.g. not just English text - biological sequences, physics data, etc). Some of our early tests indicate that this detection setup could broadly work well for any property, but we've mostly validated on hallucinations. If you have access to self-trained or open-weight models, would love to have you try out the alpha version - running it through HuggingFace!

OpenSOHO: built to manage a small number (from 2 to ~20) OpenWRT 24.10 based network devices. https://github.com/rubenbe/opensoho

I have been tinkering on a little price drop alert scraper written in Python. It's run as a cron job, and every day it checks the prices of a list of urls (my clothing staples from various outdoor retailers, mostly) and sends me an email with any products that have gone on sale.

I've been running it for over a year, but now I have fixed it up and made a little landing page to see if there's interest for a stupid-simple price watch service like this (no need to install an extension or create an account):

https://www.curiositry.com/price-drop-alert/

I've been working on a little game for my daughter to help her learn the movement of chess pieces. https://www.minichessgames.com

It's currently just a "maze" type game where you have to get to a goal square in the minimum number of moves (there are rocks placed on the board to act as obstacles)

I'm in the process of making some very simple games like battling knights where they leave poo and you try to trap your opponent.

Fun making it even if it's just the two of us who'll enjoy it :). Partly I wanted her to learn that you can create for the internet not just consume...

For the past few months, I've been building a health tracking app called LogBuddy. I got tired of using separate apps for nutrition, workouts, weight, and period tracking. So I built LogBuddy to handle all of it with a dead simple interface. That way, all the heath data I'm tracking would be in one place. Right now there's an Android APK available. If there's interest, I'll publish to the Play Store and build an iOS version too.

Would love to hear your feedback if you try it out!

Here's the github repository: https://github.com/aabiji/logbuddy

My custom poker study tools:

https://poker-study.onrender.com/

I really like the range memorization tool from GTO Wizard, but want to be able to put in custom/arbitrary ranges to test. I also want to be able to import and simplify ranges from other sites. Work in progress, but every scenario is url encoded (warning: subject to future breaking changes) and I use those urls in for links in my Anki decks.

https://github.com/nwestallen/poker-study

I suspect AI company want improved efficiencies and developing a framework that can be applied in determining the minimal-energy, maximal-efficiency architecture for ai models. Calculating the precise limits, like a Cognitive Event Horizon, where a model becomes so complicated it literally costs more energy to run than the knowledge it provides, and the Semantic Horizon, where it simply gets too complex to be accurate, etc. Lots of cool implications such as around a fundamental mathematical maximum learning rate which results in trying to get anywhere close to that that by doing stuff like aggressively filtering of the data.