I build and test
real things with
frontier AI models.
New models. Strange ideas. Things that work—and the lessons in things that don’t. This is where I make sense of it all, one experiment at a time.
Find something to read ↓On the workbench
A little hardware.
A lot to figure out.
A custom desk, a markets view, and a voice bot on an ESP-Mosaico board. Notes on the interface, the companion software, and what still needs a real-device test.
Open the build notebook ↗Notes & observations
Gemini Decisions: the question has to fit the box
Up to 1,000 bounded questions. I’d start with three—and an escape route.
↗
Gemini 4 Carbon: long jobs, small windows
A model for extended work makes checkpoints more interesting than chat bubbles.
↗
When reasoning is less visible, show the evidence
A desk UI should distinguish model narration from confirmed actions.
↗
ESP-Mosaico: a small screen for a bigger system
A custom desk, a markets view, and a voice bot. Three modes, one hardware experiment.
↗
The distance between a demo and a system
The interesting part starts when the happy path ends.
↗
A model release is a starting question
A working checklist for reading the next announcement.
↗
A voice bot needs a good silence
Listening, thinking, speaking—and knowing when to stop.
↗
No notes match that search. Try another word or choose All.
In the margins
The notebook continues on X
Smaller thoughts.
Open conversations.
Alongside the longer notes: questions about new models, ideas worth following, and updates as the work develops.
Follow @tjaichampA notebook with room to grow.
I’m following the ideas behind new models, building things with them, and keeping the useful lessons. More experiments will land here as there’s something worth saying.