GoParse

Identity verification without data hoarding

Every onboarding flow, rental contract and age check seems to end with a photo of an identity document sitting in someone’s inbox or object storage — indefinitely. We built GoParse on a simple conviction: you should be able to extract the data you are legally allowed to process without keeping a copy of the document itself.

So we made retention structurally impossible. Documents pass through our pipeline in memory, the structured result goes back to you, and nothing else remains — only aggregate metadata such as document type and timestamps, enough to run your dashboard and your invoice.

The technology

Under the API is a research-grade document pipeline, built stage by stage rather than as one opaque model. Each stage does one thing well and hands structured evidence to the next.

  1. 1

    Text-prompted segmentation

    Documents are located in the photo by text-prompted segmentation, so the pipeline finds “a passport” or “an identity card” even on cluttered backgrounds, at an angle, or several to a frame.

  2. 2

    Orientation correction

    Each detected document is cropped and re-oriented — rotated, flipped or skewed photos are normalised before any reading happens, which is where most naive OCR breaks.

  3. 3

    Vision-language extraction

    A vision-language model reads the visual zone in context: it knows the difference between a family name and an issuing authority because it understands the document layout, not just the characters on it.

  4. 4

    ICAO 9303 MRZ validation

    Machine-readable zones are parsed and every check digit is verified per ICAO 9303. When the MRZ and the visual zone disagree, the MRZ-confirmed value wins — and when nothing is confirmed, the field comes back empty rather than guessed.

What we optimise for

Privacy by design

The safest place for a copy of someone's passport is nowhere. Our architecture makes retention impossible: documents are processed in memory and discarded when the response is sent.

Honesty over guessing

An empty field is useful; a wrong one is dangerous. When the pipeline is not confident, it says so — with an empty string, a warning, and never a fabricated value.

The team

A two-person team in Udine, Italy that has spent too much time with identity documents and intends to keep it that way.

Portrait of Michele Lizzit

Michele Lizzit

Co-founder — product & compliance

Spent years reading identity documents by hand and typing the same fields into forms, and decided software should do that job — without keeping copies. Runs the commercial side and our GDPR programme.

Marco Rossi

Co-founder — engineering

Computer-vision engineer who cares more about the failure cases than the demo reel. Built the extraction pipeline, the API gateway and the dashboard — and still reviews every model change.