I’ve been away for a bit as I’ve been onboarding with my new employer and working through family circumstances. Now that things are returning to normalcy, I wanted to share an update on my different projects.
I’m splitting my attention between four projects, two of which are instructive and two of which are productive endeavors. The instructive projects are related to learning and teaching technologies that I have some surface level familiarity, as a means of improving my own skills and also sharing them with colleagues. The productive projects are already known – Medusa and Chom AI.
The instructive projects involve two different technology stacks. One is a simplistic content management system (CMS) which can be used to make blogs and websites hosted on very low power servers – think LiveJournal, or a very lightweight WordPress. (Is LiveJournal still around?… I hope mine aren’t.) The other is a proof of technology for a data processing pipeline showing how different stacks can be used complimentarily – Java/Spring stack, Elastic stack, and Podman/Kubernetes.
The CMS might be exciting in the future if it pans out — I’d prefer using that over WordPress. The constant security updates, templates breaking, lack of a working dark mode, etc… thousands of papercuts when all I want is to write some text, and maybe embed some images or code blocks. I miss the simplicity of old blogs.
For Medusa, I’ve been warming myself up on the engine again — I updated Bevy to the latest version (breaking changes are always fun), got everything working, and then implemented a tiny feature to the attract screen for fun. Start small, grow tall – valuable tree advice. Revisiting Bevy has me understanding the ECS a bit better. I’ll probably make a post related to that and how I’m going to structure the puzzle map state for Medusa in-depth.
For Chom, I’ve converted some recipe information into embeddings with OpenAI and I’ve been using that with the Postgres vector support plugin to do lookups, and that’s working well. I’ll do the same for a list of ingredients and allergens, then work on having one of Chom’s brains handle interpreting the right terms to query for semantic similarity in the database.
With the recent improvements to non-OpenAI LLMs, I’ve been considering spending some time up front making Chom run on those instead… it’d be cool to have a Raspberry Pi kit or tablet or something that runs a mini-Chom in the kitchen with a touchscreen and voice commands that doesn’t need internet to function, however far off that is for the moment. Making sure I stay compatible with non-OpenAI models will continue to be a priority either way, even if I’m not actively working on running on (say) Mixtral.
I think I previously promised a demo of Chom as well as a newsletter, which were then put on hold as the whirlwind of life swept through. I’ll reassess the timeframe for a demo release and hopefully send out the first newsletter with it. I also need to get my survey out to a wider audience.
Well, nap time before work starts. Don’t remember why I woke up so early but at least I wrote something productive as a result.
Ta.