Case study 002, 2024, fintech
Explains Spanish mortgages in plain language.
An AI assistant that explains Spanish mortgage contracts clause by clause, next to the original text, and flags unusual terms for human review.
- Role
- Lead Design Engineer for UX, prompt design, and accessibility, with an ML engineer and two full-stack devs
- Problem
- Spanish mortgage contracts average around 42 pages of dense legal text, and in our research almost nobody read them before signing.
- Decision
- The original clause and the plain-language explanation sit side by side. Every explanation cites its clause number, unusual terms get flagged for human review, and each contract carries a visible legal-verification stamp.
- Result
- In production as client work.
Context
The problem
People found out what they had agreed to when a clause took effect years later, and mortgages became one of the biggest sources of banking complaints in Spain. Customers wanted to understand the contract. They were afraid that asking questions at the signing table would make them look slow.
Decision
Never replace the contract
The assistant had to earn trust from people who distrusted banks and distrusted AI too. In the research, people named two conditions for believing an AI explanation. They wanted to compare it with the original text, and they wanted a human expert to have checked it. So the design never hides the contract, and each part answers one of those conditions. The side-by-side view and the clause citations let people compare. The human-review flag on unusual terms and the legal-verification stamp show that a person checked.
Build
How it works
I co-designed the AI architecture with our ML engineer as two prompt stages. The first extracts the contract into structured JSON with amounts, dates, clauses and parties. The second writes plain-language explanations that must cite the clause, speak to the reader as "you", and give no advice beyond what the contract says. I owned the prompt design end to end. I also designed the error flow for low-quality scanned PDFs that OCR struggles with, and ran the WCAG 2.1 AA accessibility pass that Spanish banking regulation requires.
Context
Result
The interface on this page is abstracted.
Next case
Build 4 Latam