- The assistant reads a photographed or scanned ID with intelligent character recognition and a vision-language model, then extracts every field your onboarding process checks.
- It returns a structured result of verified or flagged, with the reason, so your KYC analyst makes the final call on anything unusual.
- It flags expired documents, missing fields, and details that do not match the claimed identity.
- Enterprise h2oGPTe runs on-premises, air-gapped, private cloud or cloud, so identity documents stay inside your environment.
Every ID gets checked by hand, and some are built to pass
Your onboarding team reads each identity document one at a time: name, date of birth, expiry date, and number, compared against the application. A stolen or synthetic identity that clears a quick look becomes a fraud loss or a mule account later.
Manual review is slow for the good customers waiting in line, and it still misses documents built to fool it. Every minute spent checking a genuine ID is a minute not spent on the ones that need a closer look.
How the assistant checks a document
Enterprise h2oGPTe reads an uploaded photo or scan with intelligent character recognition and a vision-language model, then checks the result against your rules before handing it to a person.
Step 1: capture the document
Your customer uploads a photo or scan of an ID, such as a driver's licence, during onboarding.
Step 2: read every field
A vision-language model reads the image the way a person would: name, date of birth, expiry, card and licence numbers, address, class and conditions, including photos taken on a phone.
Step 3: check the fields against your rules
Retrieval-augmented generation pulls the extracted fields and your verification rules into one check, and flags missing data, expired dates, and details that do not match the claimed identity.
Step 4: return a structured result
The output is JSON: each field, whether a photo and signature are present, and a status of verified or flagged.
Step 5: route anything unusual to a person
Verified documents move on. Flagged ones go to your KYC analyst with the reason attached, and the analyst makes the call.
Where this fits your KYC process
The assistant sits beside the identity checks you already run, not in place of them. Document AI, the extraction capability in Enterprise h2oGPTe, keeps PII redaction controls on for identity data, and everything runs on-premises, air-gapped, private cloud or cloud, so documents stay inside your environment.
Proof
H2O can be used to read identity documents at onboarding, check every field, and flag anything that needs a person. No customer result for this use case is published yet.