Case Studies | FinTech

Production-grade document intelligence for carbon markets

A FinTech client

About

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

Challenge

The build focused on four areas that separate a capable prototype from a platform fit for regulated production.

A reliable, orchestrated agent chain

The prototype's four agents, a classifier to route queries, a planner to generate SQL, an executor to query the database with auto-correction and an aggregator to synthesise results, needed to run as an automated, observable pipeline rather than a manual notebook. That meant deploying each as a Lambda function, orchestrating them with Step Functions, and handling errors and retries gracefully at production scale.

Confidence-based human review

In a regulated setting, not every extraction should be trusted blindly. The platform needed confidence scoring on every extracted entity and an Amazon Augmented AI review workflow that routes uncertain outputs to the client's underwriters, with quantitative guardrails on structured fields such as credits issued, credits retired, project size and verification dates.

Citations and an audit trail

FCA requirements and EU AI Act standards demand provenance. Every extracted entity had to be linked back to its source document, page and registry through a citation layer, with an immutable, tamper-evident audit trail so any figure can be traced to where it came from.

Cost control and observability

Bedrock spend can climb quickly during intensive processing. The platform needed dashboards tracking throughput, cost per document, accuracy and review queue depth, together with spend alarms to prevent runaway costs as models and workloads evolve.

Solution

Cloud Combinator delivered the platform in phases, standing up the foundations first and layering agents, validation, audit and observability on top, so each capability could be proven before the next was added.

4

Specialised AI agents in an orchestrated chain

~5,000

Documents a month at peak throughput (indicated)

>=90%

Target classifier routing accuracy

By the numbers:

  • 4 - Specialised AI agents in an orchestrated chain
  • ~5,000 - Documents a month at peak throughput (indicated)
  • >=90% - Target classifier routing accuracy
Changes

The engagement delivered a production-grade, serverless multi-agent extraction platform against the success criteria in the Statement of Work, with acceptance measured by all four agents deployed and functional, the citation layer operational, the CloudWatch dashboard live and the A2I workflow demonstrated end-to-end. The figures below describe the design targets and indicated workload rather than an approved production benchmark.

  • Automated, orchestrated extraction — The four-agent chain runs as AWS Lambda functions under Step Functions with error handling and auto-correction, replacing manual notebook runs with a repeatable production pipeline.
  • Human review where it matters — Confidence scoring and an Amazon A2I workflow route uncertain extractions to underwriters, with quantitative guardrails on key structured fields to catch errors before they propagate.
  • Traceable to the source — A DynamoDB citation layer links every extracted entity to its document, page and registry, with an immutable audit trail aligned to FCA and EU AI Act expectations.
  • Costs under control — CloudWatch dashboards and Bedrock spend alarms give the client's visibility of cost per document and throughput, and the guardrails to prevent runaway spend.
  • Handover — The client received the platform as AWS CDK infrastructure as code with environment separation, tests, documentation and recorded knowledge-transfer sessions for its data science team.

With a compliant, observable extraction platform in production, the client can process carbon market documentation at scale with confidence, and has a foundation ready for future phases such as graph-based retrieval as its underwriting needs grow.

AWS Stack

Amazon Bedrock

For large language model reasoning across the classifier, planner, executor and aggregator agents.

AWS Step Functions

For orchestration of the multi-agent extraction pipeline with retry and confidence routing.

AWS Lambda

For serverless compute for each agent and the confidence-scoring middleware.

Amazon Augmented AI

(A2I) for human review of low-confidence extractions by the client's underwriters.

Amazon Aurora PostgreSQL

For the structured data store queried by the agent chain.

Amazon DynamoDB

For the citation and provenance index linking entities to source documents.

Amazon CloudWatch

And AWS CDK for dashboards, cost alarms and infrastructure as code across dev, staging and production.

YOU MIGHT LIKE

Related success stories

View all case studies

Case Studies | Insights

Utilising Language Recognition, Speed, and Enhanced Security to Make Social Media a Force for Good

  • Here, we take a detailed look at how the Cloud Combinator team collaborated with another cutting-edge AI service provider that provides intelligent systems to “make social media more social” for brands and users alike.
  • Arwen AI is a UK-based startup specialising in AI solutions to manage and enhance brands’ social media interactions. Founded in 2020 by Matt McGrory, Dr. David Cole, and Joel Bailey, Arwen. AI focuses on using AI to automatically detect and remove spam, toxic comments, and other unwanted content from social media platforms.
  • The team at Arwen have three core products. ‘Moderate’ is focused on identifying and removing toxic content from social media channels. ‘Engage’ helps brands identify and engage with meaningful conversations on social media, and ‘Customize’ allows brands to apply bespoke algorithms to their channels - creating an even more effective moderation and engagement.
Read more
CONTACT US

Ready to turn AI into impact?

We'll help you spot the highest-value opportunities, reduce risk around your first AI initiative, and define a clear path to results from day one.

Why talk to us:

Outcome-driven recommendations

AWS-recognised delivery expertise

Risk-aware AI adoption

Clear next step, not a sales pitch

Start with a focused 20-minute conversation about your goals — no pressure, no commitment.

We use cookies

This website uses cookies to enhance user experience and to analyze performance and traffic on our website.

See our Privacy Policy for details.