To aid my process to shortlist house for purchase I am creating a web app for this.

The app will use genai to extract the details of houses listed for sale and then update my custom database.

For example I can input a youtube url and it will fetch the transcript and use llm to generate Json response based on predefined schema.

I can review and shortlist the houses based on various custom parameters using the web interface.

The mvp is done in a proprietary tech stack, I just need to port it to open source tech stack with React and FastAPI

I'm working on a Dutch dictionary for immersion learning; https://hetnederlands.com. Structured outputs combined with react router renders a chain of thought on demand. With most "learners dictionaries", you're limited to the basic words. This enables getting a perfectly laid out learners dictionary explanation of any word in the entire language.

I got frustrated with the current web becoming more and more unusable and made a browser extension to fix that.

It's an addon that allows one to modify any web page behavior. You can write rules for things like redirecting reddit to old reddit or keeping youtube videos playing in the background.

It's like Tampermonkey/Greasemonkey/etc but it works on HTTP request level and can modify scripts/styles/markup before they are ever rendered making it considerably more powerful.

Currently it only works on Firefox (and Firefox Mobile) because it's the only browser that supports the necessary APIs. It's seen its first release already but the UI is a bit lacking :)

AMO: https://addons.mozilla.org/en-US/firefox/addon/web-defuser/

Sources: https://gitlab.com/gear54/web-defuser

Example rules: https://gitlab.com/gear54/web-defuser/-/snippets

Have been building a design tool for Spatial Computing [0] Me and my co-founder struggled for over a decade working in XR and wanted to simplify the design process for us and everyone else. We've been focused on reducing steps between design iterations by as much as humanly possible. So the tool works live, multi-user, in real-time and immediately across all compatible devices.

Here's a recent example of whats possible from one of our users [1], and a recent project I made for fun that you can try on your iOS [2] or Meta Quest [3] device.

[0]: https://ordinary.space

[1]: https://www.linkedin.com/posts/matt-rice-sennep_mixedreality...

[2]: https://app.ordinary.space/#/share/b2d3d7

[3]: https://app.ordinary.space/#/share/96b9cf

Writing a book on my unique theory on human psychology based on cynicism, military theory, and evolution.

Is has helped me to address all of my problems, from mental health to productivity to learning across all domains like economics, writing, coding, marketing.

Writing daily blog posts, and then piecing together everything into the book. https://www.moderncynicism.com/

I have a good understanding of how to create likely-correct software: https://www.osequi.com/studies/list/list.html

Now I'm learning AI/LLMs from the perspective of correctness. So far I have two 'maxims' to guide me:

- AI shines where humans struggle (for prompt engineering)

- An LLM is nothing but an API call (for software engineering with AI)

Working on an embedded DSL for writing typed LLM workflows in Haskell: https://github.com/outervation/promptyped . It handles the "LLM in a loop" structuring for e.g. forcing an LLM to keep making changes in a loop until unit tests pass, and custom validation so the LLM is forced to retry a task if the validation fails (validation can include using a LLM to check the result, or plain Haskell function, or e.g. running a compiler/unit tests). It also allows structured context management, allowing the user to specify which files are initially provided as context to a task (which may itself be the result of a LLM query), as well as providing the LLM tools to open files it deems relevant, and prompting it to close the least important file when it has too many open (and compressing non-recently-modified source files when many are open).

Currently heavily WIP though and only has a working Go compile/test backend (Go seems particularly easy for smaller LLMs like o3-mini to work with due to being a relatively non-complex language).

We are working on a fast reading font which works by using just opentype features (no server or other script required)

https://github.com/Born2Root/Fast-Font

We also created some emoji fonts: https://github.com/Born2Root/Feature-Fonts

Busy building some personal infrastructure around information. Think mix of scrapping and filtering, scoring and summarization.

Clearly all of the info space is getting ever more polluted and I just don’t trust anyone else (with their own agenda) to manage and filter that for me. If one is to abdicate that sort of responsibility to a system wholesale then I think it has to be fully under one’s control, own data, own design, self hosted etc

I’m wrapping up the final audio plug-in in my “Loudness Series” which was designed to bring compressors and limiters into modernity by upgrading their circuitry to using LUFS and True Peak in place of RMS/Peak.

The final plugin is an equalizer which will be focused on loudness but expand the filtering to include more advanced equal loudness contours.

https://apu.software