501–510 of 1001 posts
[0]: https://sotix.itch.io/plight-of-the-wizard/devlog/1517881/v0...
Unfortunately the only marketing material so far are some TikTok posts, but it's a pretty cool demonstration.
I’d love any feedback!
It’s a lot of fun and ultimate nerdery for me :) I’m a ski instructor through the Austrian and Danish ski school systems, I studied physics, and I’ve been a developer the last 15 years.
The original developer has left our organization so I’ve been tasked with general assessment and winding it down to enter maintenance mode. It’s still alive and well, has a very passionate and appreciative user base, but we want to ensure it doesn’t demand too much attention moving forward while we focus on other things. It has pretty noisy error reporting.
Reliability and fault tolerance are some of my favourite things to work on in software so it has been a lot of fun so far. It has also been an incredible opportunity to practice using LLMs for specs, planning, verification, and research. I don’t actually need to output much code to get this thing into a stable state in which it can coast along; the bulk of the work is time spent understanding the app, the infrastructure, its existing faults, poring through traces and logs, going over query plans, and so on. LLMs are great assistants for this work and I’m having a ton of fun having so many opportunities to figure out what works and what doesn’t.
The outcome has been awesome. The performance is steadily climbing (especially in the database), and most common errors when I started are either gone or much better understood with plans to address them. I’ve almost got it set up so if someone needs to take it over in the future, it should be pretty easy to toss them the keys and trust that they can deploy and maintain it easily from the docs and systems I’ve created.
Despite spending a lot of my career on the front end, the hardest part of this project has been navigating that. Aiming to improve an application with minimal intervention is exceedingly difficult in the browser, or so I’ve been finding. I can get incredible performance gains out of Postgres without changing the interface between lambda and rds in the slightest, but meaningful improvements to the react application seem virtually impossible without substantial refactoring.
I understand the key factors in getting better performance out of react apps and I see plenty of opportunities, but they all involve large diffs that are risky and time-consuming, even with a model like Opus handy to churn through boring and large change sets. It’s such a fragile and flaky environment.
Even so, I’m loving it. Making software better is so gratifying. Doing it without reinventing the world is such a fun challenge, too. It really puts your brain to work. It would be so easy to go in and start flipping tables and throwing code in the garbage, but that’s too easy and too risky. Taking it slow, absorbing as much information as you can, truly understanding how features work, and planning surgical changes with significant pay off is safer and just feels awesome when it works.
I’ll be sad when this one is finished. It’s almost there. Next up is a remote temperature controller for 40 saltwater experiment tanks with a temperature profile planning interface and a monitoring interface for the lab. That will be awesome too. It has been a good couple of months for work.
595 days and counting.
ᕙ(⇀‸↼‶)ᕗ
- Calculator that gives tg/s and vram required based on model params and ddr settings.
- Auto create dashboard from csv/json files or apis Easyanalytica.com
- snippet viewer for html/react that allows annotation and sharing based on url fragments
- Integrated with lots of open-source and commercial simulators and models for chemistry, materials science, biology… As well as connections to service labs and robot labs to easily perform physical experiments.
- autoresearch / AlphaEvolve like optimization loop following the scientific method: observation, hypothesis, experiment, theory. Combined with a long-term self-learning memory like Karpathy’s Wiki.
You can work with it interactively like with a coding agent to research and execute experiments efficiently. You can also treat it like a graduate student, giving it long-term research goals, having it work 24/7, making smart decisions about where to use your limited resource budget, checking-in with it periodically as a supervisor to guide its direction.
Not all of this is shipped yet, but we’ve been online for a while and it should be plenty useful to any scientist/engineer already.