area · AI Engineering
AI work with an artifact trail: spec-first Claude Code workflows wired into the issue tracker, and product work on an LLM document-extraction platform.
◆ ExtractionDemo
Press «Extract»: the JSON builds up field by field, the way it arrives from the model, and each field highlights the part of the document it came from. In production, that traceability is what makes a verdict auditable.
POLICY REQUEST · MEDICAL EXPENSES INSURANCE
Applicant: Ana Ejemplo Ruiz
RFC: EERA850101XX0 · Date of birth: 01/01/1985
Address: 12 Fictional St, Downtown, ZIP 00000
Declared occupation: Software engineer
Requested sum insured: MXN 2,000,000
extraction.jsonwaiting
Jira ticket
The task arrives with its requirements template and technical context.
Claude Code
Works on the repo following the specification (spec-first).
Validate docs
Checks the ticket's documentation requirements in the repo.
Close ticket
If everything is in place, marks the ticket as complete.
Demo with fictitious data: no real clients and nothing from production.
LLM document extraction in production inside an AML/KYC module, with an auditable verdict for each check.
Read the case study →Wired into the issue tracker: it validates documentation requirements in the repo and closes tickets.
An endpoint-by-endpoint review of a 339-endpoint legacy backend, cross-checking every gap before declaring it.
End-to-end owner
AML/KYC module of an insurance platform (art. 492): it ingests a request's attachments, classifies them, extracts their fields, runs checks and stores a verdict with the reason for each one.
I reply in English or Spanish.