Live demo · synthetic data
Banking
compliance agent
An AI that reviews loan agreements against the FCA Consumer Duty, cites the regulation behind every finding and hands anything uncertain to a person. Below: what it found in 34 test agreements, how reliable it was, and a step-by-step replay of real reviews.
This page is static and always on. The interactive review app runs on Streamlit and may take about 30 seconds to wake if nobody has used it recently.
01 · Compliance overview
What it found, and how reliable it was
Findings by Consumer Duty outcome
Each agreement counted once (its first review), as the reviewer receives it.
What the four outcomes mean
- Price and value
- Is the price fair for what the customer gets? Fees and charges must be justified.
- Consumer support
- Can customers get help when they need it, especially vulnerable customers?
- Products and services
- Is the product designed for, and sold to, the right customers?
- Consumer understanding
- Are the key terms explained clearly enough for the customer to understand?
How reliable is it?
Vulnerable customers
System health
02 · Try a sample
Follow one agreement through its review
A replay of the real recorded run. The code rules are the project's own, run on this data. Nothing here calls a model.
Read the agreement
03 · Be the reviewer
Every case goes to a person
Open the review queue in the interactive app. Each verdict shows its reasoning, the quoted claims and the full FCA passage. Confirm it or override it. An override needs a written reason, and nothing resolves until every item is decided. Those rules are enforced in code.
Hosted on Streamlit Community Cloud. If it is asleep, press “wake up” and wait about 30 seconds.