Job Description
Seeking a Computer Vision Engineer to detect altered and fabricated documents across 20 document types. You'll build Python-based checks for images and PDFs, measure how reliably they detect tampering, and give fraud reviewers clear evidence for each flag.
Location: Remote (US), Sacramento or nearby preferred
What you'll do
- Build checks for inconsistent fonts, text alignment, image edits, and anomalies in PDF structure or metadata.
- Create a labeled evaluation set using genuine documents, confirmed examples of tampering, and synthetic alterations. Test against emerging methods, including AI-generated documents.
- Tune detection thresholds with fraud reviewers. Measure precision, recall, and false positives at realistic fraud rates and across different image-quality levels.
- Run new checks in shadow mode before they affect review decisions, monitor performance after launch, and show reviewers which region or file detail triggered each flag.
Qualifications
- 3+ years for mid-level or 6+ years for senior in applied computer vision, machine learning, or fraud detection, including at least one model deployed to production.
- Hands-on Python experience with OpenCV or scikit-image, plus PyTorch or TensorFlow.
- Experience inspecting images and PDFs programmatically, including rendering, metadata, and file structure.
- Experience evaluating models on imbalanced datasets, selecting thresholds, and explaining findings to nontechnical reviewers.
Experience in document or image forensics, identity verification, or claims fraud is preferred. Familiarity with PyMuPDF, pikepdf, or ExifTool is a plus.
Share your resume and one brief example of a detection model or check you personally built. Include what it detected and how you measured false positives.