Skip to content

Recipe: anonymize a CV with a photo

A CV is a classic mixed document: identifying text (name, email, phone, address, a personal website, dated job history) and an identifying image (the headshot). Piixie handles both in one pass.

Goal: a CV a hiring panel can review blind — no name, no contacts, no recognizable face — that still reads like a real CV.

Use synthetic, not redaction: a blind CV still needs to read naturally so reviewers can judge the experience. The default local model is fine; bump to reasoning Low/Med so the dated job history is handled thoughtfully.

Drop the PDF in and process. Because it’s a PDF with an image, Piixie’s vision pass notices the headshot. The run produces the usual entry list plus the photo as a flagged image.

Open the result in the editor. Expect entries like:

TypeOriginal
NAMEMarcos PatelDavid Romero Gil
EMAIL[email protected][email protected]
PHONE+34 612 345 678+34 698 112 530
ADDRESSC/ Mayor 12, MadridAv. del Sol 84, Sevilla
URLlinkedin.com/in/mpatel
DATEJan 2020 – Mar 202302/2018 – 11/2021

Check the dated job spans got caught — they pin down a timeline (why dates matter). If one slipped, add it by selecting it.

In the entries panel, find the Images section with the headshot thumbnail. Flip its blur switch on. The regenerated CV now has an unrecognizable photo.

If the photo is part of a flattened scan rather than a recognized embedded image, draw a Black box over it instead.

A blind CV usually keeps employers and schools (they’re relevant to the decision) but scrubs the person. If the model synthesized a NAMEd reference or a former manager you’d rather keep generic, that’s fine; if it over-scrubbed the well-known employer, turn that entry off so the company name survives.

Save to output. If you process CVs regularly, save a template now — synthesize NAME/EMAIL/PHONE/ADDRESS/URL, keep employers, treat spans as DATE — so the next CV is one click. (The photo’s blur is per-file; templates don’t carry image or drawn decisions.)

A natural-reading CV with a fake identity and a blurred face — reviewable on merit, with the candidate’s identity left on your machine.