Candidate CVs into the firm's house format
The problem
Consultants were rebuilding every incoming CV into the firm's own template by hand — hours per candidate, and no two consultants produced quite the same document.
What we built
A pipeline that reads the candidate's CV alongside the consultant's interview notes and the record already held in their CRM, fills a fixed house template from all three, and renders a finished DOCX and PDF for a consultant to review field by field before anything reaches a client.
The decisions that mattered
- A field is never blank and never invented — anything the source documents do not answer is marked as pending a human, not guessed at
- Signed off is a status only a person can set; nothing the system does will set it, and it refuses while a required field is unresolved
- Every value shows which document it came from, and anything left out shows why — reversibly
- Corrections teach it. Its own output never does, so a mistake cannot harden into a rule
- Candidate-side and client-side staff see different data, enforced on the server rather than in the interface
- A candidate can be erased completely — the record, the rendered documents, and the original uploads they were built from
Where it stands
The extraction engine has been proven end to end on real candidate CVs on the live account. The review, access and delivery stages around it are built and being put into daily use.