Document extraction & assessment
Process supporting documents into structured information for review, comparison and downstream loan-administration work.
Back to work
AI-assisted bank disbursements
Refined document-extraction prompts, challenged AI workflows to uncover weaknesses, and became a key point of contact for disbursement and progressive financing flows at a major bank.
My roleAI engineer · Disbursement & progressive-flow PIC
Overview
Loan operations connect document evidence, business rules and staged disbursements. This project combines AI-assisted document processing with a portal for reviewing and progressing loan-administration work.
Project statusEnterprise project delivery
Scope
An AI-assisted loan-administration platform for a major bank. The wider system covers application information, supporting documents, review checklists and disbursement workflows. My role centres on extraction prompt refinement, testing for weaknesses and responsibility for key disbursement and progressive-flow areas.
Main features
Disbursement workflows
Support the preparation and review of disbursement information, connecting document evidence with the relevant application and checklist context.
Progressive financing
Handle subsequent disbursement stages, where claims and supporting information need to remain connected to the wider loan workflow.
Operational review portal
A web portal brings application details, document batches, workflow status and audit information into the operations team’s working environment.
Technology
React & TypeScript
Typed web interfaces for the operations portal.
TanStack Query, Router & Table
API state, application navigation and tabular operational views.
Material UI & Tailwind CSS
Interface components and styling in the host application.
Python
Backend processing and loan-administration APIs.
Amazon Bedrock
Model inference for document extraction, assessment and transformation.
Lambda & API Gateway
Serverless processing and the portal’s API layer.
DynamoDB & S3
Application state and supporting document storage.
OpenSearch
Search and filtering for operational listings and audit information.
Microsoft Entra
Enterprise identity integration for the administrative portal.
The stack describes the supporting platform, not sole authorship of every component. Client-specific integrations, internal identifiers and operational documents remain private.
Engineering approach
Prompt refinement against document evidence
The extraction work focuses on how document content becomes structured information that the workflow can use. My contribution includes refining prompts to improve that interpretation.
Challenge the happy path
Adversarial testing and difficult input scenarios help reveal weaknesses that a successful demonstration can miss.
Connect extraction to the operational flow
An extracted field is only useful when it reaches the correct review and disbursement context. My responsibilities expanded from AI work into key disbursement and progressive-flow areas.
What I worked on
- Refined prompts for document extraction as an AI engineer.
- Used adversarial testing and challenging input scenarios to identify weaknesses in AI-assisted workflows.
- Became one of the points of contact responsible for the disbursement and progressive financing flows, working with the wider delivery team.
Results & scope
AI + workflow
extraction, disbursements & progressive financing
- Contributed to document-extraction refinement and flaw discovery within an enterprise banking project.
- Took responsibility for key areas of the disbursement and progressive financing workflows.
Part of Axrail’s enterprise delivery. My responsibilities covered AI extraction and key disbursement workflows within the wider project team.
Another project
KIP.ai
Contact04
Say hello.
For AI engineering roles, collaborations
or a conversation about further study.