Methodology
The portfolio is built around a simple rule: use automation where it clarifies structure, and keep accountable judgment with people.
Kyrantis tools use AI for first-pass extraction, coding, drafting, and consistency checks where the work is repetitive, pattern-heavy, and reviewable. They do not treat model output as the final authority. The prototypes show where model assistance could sit in a production workflow and where accountable review gates would be required before publication or formal use.
The public versions are static prototypes by design. They are free to host, private by default, stable for review, and have no live keys or API calls to leak. That constraint is not a workaround; it is part of the product method. Each tool can show the policy workflow, data boundaries, interface structure, and expected outputs without sending public visitors' text to a model provider. Where a live system would use a model, the landing pages now specify the production architecture and the review gates that would sit around it.
The public case libraries link back to their sources and label evidence-confidence tiers, but the records have not yet been independently double-checked entry by entry. That distinction matters: a traceable prototype can be useful for portfolio review while still making clear that each entry would need independent verification before anyone relies on it formally.
The case libraries are intentionally small. The Atlas contains 12 documented deployments and the Incident Tracker contains 8 landmark incidents because the portfolio prioritizes per-record detail over large, unverified aggregation. A smaller set can preserve source links, limitations, evidence-confidence labels, risk rationale, safeguards, and policy lessons in a way a large scraped database often cannot. The goal is not to claim coverage of every public-sector AI system. It is to show how a sourced, reviewable policy evidence layer could be built and governed.
Return to policy tools