LLM experiments
Open models, prompt patterns, retrieval pipelines and evals: what works, what doesn't, and why.
pocbox is a sandbox for LLM experiments, agents and AI tooling. Every idea gets a fast build, an honest test, and permission to break.
Not production. Not a portfolio. A place to test whether an idea actually works before it deserves more.
Open models, prompt patterns, retrieval pipelines and evals: what works, what doesn't, and why.
Tool-using agents, bots and scheduled workflows that do real work, wired to real APIs.
Small developer tools and APIs with a model inside, spun up in minutes, tried for real, and torn down without ceremony.
The whole loop is designed to be boring, so the experiment is the interesting part.
Pick one question an idea has to answer.
Ship the smallest thing that answers it. Prompts first, code second.
Test it against real examples, not vibes. Note where it breaks.
Promote it, iterate on it, or delete it. All three are wins.
The questions worth a prototype right now. This list changes as the field does.
We're not saying what it is yet. It's being built, tested, and broken on purpose. We'll reveal it soon.