AI engineering / verifiable systems
Texts about AI and crypto
Notes and essays on AI engineering, crypto, proof systems, and product work.
I’m Omar U. Espejel. I work on AI and crypto at Starknet Foundation. Before that I worked on crypto at StarkWare and machine learning engineering at Hugging Face.
Latest essays
-
Calibrating the SO-101 Is Hardware QA
A practical SO-101 leader/follower calibration walkthrough, with measured ranges, benchmark comparisons, encoder-wrap failure modes, and the no-motion checks to run before teleop.
-
The Environment Contract Is Part of the Product
A postmortem on an auth flag regression in an agent-built app, and the deploy contract that stops stale environments from becoming releases.
-
What a Proof Is Allowed to Mean
A guide to the missing receipt around proof artifacts: how proof bytes, statement meaning, verifier domain, and replay assumptions become one accepted object.
-
Proof Validity Is Not Statement Validity
A proof can verify while the AI claim around it is still relabelable. In a local EZKL-style receipt test, the proof-only path rejected 1 / 7 relabels; the statement-envelope path rejected 7 / 7.
-
Proof Pressure Is Not Just Matrix Multiply
The useful transformer-proving question is not whether the model contains arithmetic. It is where proof plumbing gets reused or repeated.
-
Why Transformers Fit STARKs
Transformer decode is a repeated state transition over carried context. That shape does not make STARKs win automatically, but it makes the proof-boundary question natural.
-
Why Documentation Structure Fits Agent Workloads
How curated manifests, canonical metadata, crawler boundaries, and simple health checks make documentation easier for AI agents to retrieve and cite.