Computer vision developer — Engineering drawing registration and reconstruction

Person Spec We are looking for a developer with demonstrable computer vision experience for a short term project with scope for longer engagement for the right candidate. Would prefer someone with CV aligned qualifications but will expect to see related previous project work along similar lines so experience is our core requirement here. Goal Build a new Python 3.11 library, or Linux-containerisable equivalent, that reconstructs a full technical drawing from overlapping PNG tiles (hand-scan-like degradation). Feature data only — no layout files, manifests, or ancillary data. Data and success criteria We supply a simulated dataset only (no production scans). Grids: 1x4 and 2x4; each in clean and hand-scan variants. Per scenario: tiles/*.png, reference.png (evaluation only), README (overlap 12%). Guideline: 1x4 @ 200 DPI in under 2 minutes on a reasonable laptop. Project success = passing all four scenarios above. Human review will be needed to assert correctness but image registration measures would be considered if suggested Suggested stack: NumPy, OpenCV, SciPy, scikit-image (others if justified). Apply with Document/scan mosaicking portfolio; topology inference approach; registration model for flat scans; photometric handling; timeline and quote.