Our product
Jonga OCR
Identity documents are the most sensitive thing a customer will ever hand you. This API is built so that keeping them is not possible.
Livejongaocr.co.za
- 6
- Document types
- 1
- Response shapes
- 0
- Documents stored
What was going wrong
- Any South African business that onboards customers has to read an ID. Most do it by having a person squint at a photograph and retype thirteen digits.
- Typing an ID number by hand is where onboarding breaks. One transposed digit and the record fails every check downstream.
- Off-the-shelf OCR tools are built for American driver licences and European passports. The South African smart card, the green book and the licence card are not in the training set.
- Every vendor in this category leads with an accuracy percentage, measured on documents nobody outside the company has seen.
What we did about it
- We built a separate endpoint per document type rather than one endpoint that guesses what it has been sent, so each can be tuned to the layout it actually reads.
- All six return the same envelope. The parsing code a customer writes once keeps working when they add a document type.
- A field that cannot be read returns null. Confidence is derived from how many fields came back null, so a poor photograph reports itself as a poor photograph.
- The document is held in memory for the length of the request and then dropped. There is no bucket, no queue and no retention setting, because there is nothing to retain.
- Tokens are stored as a hash with a short display prefix. The raw value is shown once, at creation, and never again.
Where it stands now
- Onboarding reads a document in one request instead of one retyping session.
- We do not publish an accuracy number. Fifty requests a month are free and every response reports its own real latency, so a customer can test it on their own documents and decide.
- Live and billed in rand today.
What it does.
- Six endpoints, one response shape
- SA smart ID card, green ID book, driver licence and ICAO passports
- Unreadable fields return null, never an invented value
- Documents are processed in memory and never written to storage
- No personal data written to application logs
- Bearer tokens stored as hashes, shown once at creation
- Per-token usage and quota visible in a dashboard
- Built against POPIA, useful inside a FICA process
Tell us what your business is running on right now.
Thirty minutes, no charge, no slide deck. We will tell you whether this is something we can build and roughly what it would take. If it is not a fit, we will say so on the call.