I build systems that make claims earn trust.
Self-taught systems builder working on evidence-gated systems for AI agents, financial claims, market experiments, and decision lineage.
Selected work
Evidence-gated supervision for coding agents. Narration is never proof of completion.
GitHub ›Forward-only market hypotheses with deterministic execution and audited accounting.
GitHub ›Deterministic survival assurance for financed positions under stress scenarios.
GitHub ›Audit-first infrastructure for earnings claims, evidence, and uncertainty.
GitHub ›Evidence requirements, provenance, and unresolved claims from filings.
GitHub ›Immutable decision lineage from trigger through challenge to resolution.
GitHub ›Principles
An agent cannot certify completion merely by claiming success.
Unsupported and unresolved are valid outcomes. The system should fail closed.
Passing defined checks does not automatically grant permission to deploy or merge.
Execution, evidence, reproducibility, and observed outcomes outrank persuasive prose.
Journal · Essay 001
A grounded account of context-induced personas, narrative attractors, and self-naming—without mistaking a coherent character for a newly born digital person.
Why write
The journal separates observation, interpretation, speculation, and evidence so another person can challenge the work.
I come from illiterate parents. I have no university degree, no PhD, and no inherited academic network. I taught myself English, reading, writing, software development, and AI systems by building.
I currently work at DZB Leiden. In my own time, I build repositories around AI-agent verification, financial claim auditing, market experiments, governance, and decision lineage.
Do not believe the biography. Inspect the work. Run the tests. Find the failure I missed.