Open tools

Projects for banking AI evaluation.

Public, defensive tools I publish for teams that need to evaluate an LLM before it touches production banking workloads — not attack demos.

Live · Hugging Face Space

LLM Red-Teaming Banking Severity Scorer

Before go-live, a bank needs a shared language for how bad a finding is — not a pile of redacted screenshots. Severity has to map to finance risk and to the evaluation checklist, not to a demo scoreboard.

I built an interactive severity scorer that walks a defensive finding through impact, likelihood and control gaps, then returns a severity band you can put in a go-live review. It sits next to the article LLM Red Teaming for Banking: A UAE Go-Live Checklist.

Use it to score findings from a defensive evaluation — severity bands for go-live review, not an offensive toolkit.

Stack: Gradio · Hugging Face Spaces · defensive severity bands · UAE banking go-live context

Open the Space → Read the checklist article

Coming soon · week of 26 Oct 2026

Bank LLM evaluation checklist

A reusable checklist for evaluating an LLM before banking go-live: scope, harms, multi-turn probes, severity, re-test after every model update. Defensive only — what to verify, not how to attack.

I am publishing this as a public GitHub repository under my own name the week of 26 October 2026. This page will link the live repo URL as soon as it is public. Until then there is no placeholder link.

Companion to the severity scorer and the Hugging Face article — one checklist a risk and architecture team can run together.

Stack: Markdown checklist · GitHub · defensive evaluation · banking go-live

Repo URL — not public yet

Building AI for a regulated environment? Let's talk architecture.

Contact →