MF-01 · Finsfera · Dec 2024 → now · Fullstack Lead
AutoKYC
An identity check that fintechs and exchanges plug into. A partner sends a photo of a document and a selfie; AutoKYC confirms the person is who they claim to be, or hands the case to a person for manual review.
Team Fullstack Lead (me) · Fullstack Middle · BA · PM · CTOMy part the core verification flow, from scratch
~96%of applications decided without an operator
60+countries' documents read
10+sanctions & wanted lists screened
01 · The problem
Trust a stranger, in seconds
Before money moves, a fintech has to know who its customer is: KYC, know your customer. Checking by hand is slow and expensive, documents from 60+ countries all look different, and one name has many spellings. The goal: let the machine decide the clear cases and send only the doubtful ones to a person.
02 · What I built
What I built
Documents from 60+ countries
One reader for documents from 60+ countries: text (OCR), the machine-readable zone (MRZ), QR codes and barcodes.
See how it works
One reader for 60+ countries' documents.
Illustrative
A scan line passes over an abstract ID card; its text lines, QR code, barcode and machine-readable zone light up in turn, while a field of dots, one per country, fills in.
60+countries' documents
One flow, mostly automatic
The core verification flow, built from scratch: document reading, face matching and duplicate detection in one flow, so the machine decides the clear cases.
See how it works
Most applications never wait for a human.
Illustrative
A stream of applications: about 24 of every 25 go straight to an automatic decision; one peels off to a person.
~96%decided without an operator
Checks that run in parallel
Independent checks run at the same time, and an encryption bottleneck is gone.
Sanctions and wanted-list screening
10+ lists (OFAC, UN, EU, Interpol, FBI), matched even when a name is spelled differently.
See how it works
Every spelling lands on the same name.
Illustrative
Three spellings of a fictional Ukrainian surname, Гнатюк, Gnatyuk and Hnatjuk, converge on one name: Hnatiuk.
Two more services
Age estimation and content moderation, on the same flow.