AI ID-Scanner & Automated Onboarding

Registration flows are full of manual typing. This build — 2nd place at the HSLU hackathon (Netzwerk Neubad, Lucerne) — removed it: photograph an identity card, and the data is extracted and dropped into the web form automatically, error-free. The brief was the Autofill Form Capture challenge, set by xappido AG at the Young Talents Hackathon 2024.

The deliberate constraint: zero-cost, open source#

The interesting decision here was architectural. We could have called Microsoft Azure Computer Vision and been done. Instead we chose a free, local, open-source pipeline on purpose — to prove maximum independence and cost efficiency, and because owning the pipeline meant we understood exactly why it failed when it did.

The computer-vision pipeline#

1. Pre-processing with OpenCV. Before any text is read, the image is cleaned up — contrast optimisation, cropping, noise reduction — because OCR accuracy is won or lost at this stage, not at the recognition stage.
2. Text extraction with Tesseract OCR. The cleaned image goes through Tesseract for precise text extraction from the ID document.
3. Algorithmic data mapping. Tesseract returns unstructured text; we wrote the parsing logic that analyses that output and maps it to the right form fields — name, date of birth, ID number — which is the unglamorous part that actually makes auto-fill trustworthy.
4. Real-time front-end validation. A minimal Angular interface validates and presents the extracted data live, so a human can catch an OCR slip before it becomes a wrong record.

Stack: Python · OpenCV · Tesseract OCR · Angular.