Reading 100-page loan documents in minutes, not hours

A FinTech business working with Cloud Combinator on AWS. The client is anonymised at their request.

The client brought three connected challenges to the engagement, each rooted in the volume and complexity of the documents flowing through origination.

Cloud Combinator structured the work so that value was proven early and then hardened into a deployable service. The proof of concept validated the core pipeline against the client's real documents; the initial-works build then added the secure networking, orchestration, monitoring, and documentation needed to run it reliably. The build itself progressed through four phases.

Cloud Combinator delivered a fully functional, serverless document-processing pipeline that meets the project's success criteria: automated ingestion, Textract-based extraction, and Bedrock-driven analysis and origination decision support, all on a scalable, cost-efficient architecture, and accepted on sign-off against those criteria by the client's technical and business stakeholders.

• Automated end-to-end processing — Documents move from upload to extracted text to AI analysis with no manual intervention, removing the manual reading bottleneck at the front of origination.
• AI-assisted origination decisions — Amazon Bedrock with Claude (3.5/3.7) turns dense, unstructured PDFs into concise page-level analysis that supports faster, more consistent decisions.
• Scalable and cost-efficient by design — A fully serverless architecture scales automatically with demand, with no upfront cost and a projected 12-month AWS spend of approximately $24,302 (USD, indicative).
• A foundation for future growth — The architecture is ready to accommodate optional enhancements such as SNS notifications, human review via Amazon A2I, or a front-end interface, without rework.
• Handover — The client received source code, CloudFormation templates for reproducible infrastructure, detailed technical documentation and user manuals, and recorded demonstration sessions for a clean transfer of ownership.

With the core pipeline proven, accepted, and running in the client's own AWS account, the business now has a scalable platform to build on. The natural next steps, scoped separately, include stakeholder notifications, human-in-the-loop review, a front-end upload and review interface, and a Well-Architected review to take the service to full production standard.