Customisable blueprint
Codex DocAI
A customisable blueprint for AI document processing and KYC: capture, extraction, a human review queue and a decision log.
Overview
What this product is for
Codex DocAI is a blueprint for turning identity documents and business paperwork into structured fields a workflow can use. The model proposes a reading. A reviewer confirms or corrects it. The system keeps both. It is aimed at KYC packs, onboarding files and back-office document queues. It is not an autonomous identity decision, and it is not a claim of a particular model accuracy.
Onboarding, compliance and operations teams that still retype passports, trade licences and bank letters into another system.
At a glance
Capabilities
What is included
- Capture from a phone camera or an uploaded file
- Field extraction for the document types you define
- Confidence shown next to the field, not hidden
- Human review queue for low-confidence or high-risk files
- Side-by-side original and extracted values
- Decision log: accepted, corrected or rejected, and by whom
- Retention and deletion rules agreed before go-live
Stack
Technology
Questions
Questions before a project conversation
Does the model approve a customer by itself?
No. Extraction fills fields. A person accepts, corrects or rejects. High-risk files are designed to stop in the review queue.
Which documents can it read?
The set is defined per engagement: passports, national IDs, trade licences, bank letters and similar files. A document type that was not in scope should fail visibly.
Where are the files stored?
In object storage the client controls, with a retention period agreed before launch. The blueprint does not assume a public bucket.