Software that shows its work.

Autonomous research pipeline

A Hermes Agent research loop that screens European equities, backtests strategy hypotheses with costs, and runs a final gate before any paper rebalance - leaving an inspectable evidence trail at every stage.

Caught a dispatch bug producing clone-like backtests, refused strategies its engine couldn't test honestly, and held calculated orders when the active strategy breached its drawdown limit.

En Minutes

A bilingual data visualization app that turns French prices into minutes of work, with historical data, salary references, and methodology checks built in.

Reframes French purchasing power in minutes of work, with salary references and methodology choices visible.

Omaha Insights

A finance analytics web app for KPI monitoring, peer comparison, valuation work, and investment screening.

Makes KPI monitoring, peer comparison, valuation, and screening workflows easier to inspect and trust.

OneCode Cloud

A cloud platform for deploying Python applications from the OneCode ecosystem, built with Next.js, AWS serverless services, and a secure workspace model.

Turns Python workflows into deployable applications with workspace boundaries, permissions, logs, and cleanup.

The bug is not the bug

Whoever defines the bug controls the investigation. The first plausible explanation is a trap.

The correction is the problem

TDD biases toward testable code. Code review biases toward reviewer taste. Every fix introduces its own distortions.

Telling an LLM to 'be creative' is one of the worst things you can do

What Arthur Koestler's bisociation theory suggests, and how I ported the Open-Collider method into a Hermes skill.

When mocks lie

A concrete integration bug from an AI-built research system: both sides of a contract had passing unit tests, but the real components broke when they met.