What they did
From September 2024 to September 2026 Imran was the AI workflow engineer and product manager at Ultrastrategy, a French platform that sources C-level consultants. Senior CVs arrived in different languages and layouts, and most sourcing time went on reading and lining them up. He built n8n workflows that sent each CV to Affinda for parsing, used OpenAI models to extract the fields specific to the product, stored the profiles in a SQL database and ran a matching algorithm written against the product brief. A person made the final choice. He also built separate pipelines that generated invoices, timesheets and expenses with PDF.co.
How it went
Sourcing a C-level consultant went from about five hours to one, an 80% cut, while the platform grew from zero to more than 1,300 users. He gives no figures for parsing accuracy or cost. The rest of the post is how he would rebuild it today: strict structured output from the model, validation with a review queue for failures, hard filters in SQL before scoring, and stored reasons for each match.
Worth knowing
Structured output guarantees the shape of the data, not that the values are right, so validate every field. Under the EU AI Act, systems that filter or rank job candidates are high risk, with the rules for them now due on 2 December 2027, so build in human review and logging. Check where any parser hosts CVs and how long it keeps them.