Architectures
Interactive 3D walkthroughs of AI system architectures I've built with or think about a lot. Drag to rotate, click to inspect.
-
Context engineering · the context pile
An AI agent re-reads its entire working memory every single turn. Build one, race it against a typical agent, and see what three habits (helpers, reading less, notes) actually buy you.
Open walkthrough → -
The agentic AI stack
Three planes, nine layers. Data (substrate), Control (reasoning), Governance (oversight). Click a layer to see why it matters and what it directly touches.
Open walkthrough → -
Looped transformer inference
How a recurrent-depth transformer runs the same block R times per output token. The loop, the halt gate, and where the reasoning actually happens.
Open walkthrough → -
Laya · open-source System One decisions
Convai's 421M-parameter Apache-2.0 non-autoregressive decision model. Reuses ModernBERT as a bidirectional encoder, adds a 2-layer decision head trained with RLCD. Runs on your laptop.
Open walkthrough → -
Jev · TypeSafe's closed alternative
Same category as Laya, different architectural bet — a custom masked causal transformer with parallel sampling. Hosted API only, no weights. See Laya first; open this for the closed-source counterpart.
Open walkthrough → -
Orchestrator-worker (coming) coming
The multi-agent research pattern — one coordinator, many parallel workers, synthesis at the end.