Services provided
- Artificial Intelligence
- Data & Analytics
- Cloud Engineering
- Serverless Architecture
Platforms used
- AWS (Amazon Bedrock, AWS Lambda,
AWS Step Functions, Amazon API
Gateway, Amazon S3, Amazon DynamoDB,
AWS CDK)
Engagement length
- 12 weeks
Other stats
- 96.26% validation accuracy achieved on production validation datasets
- Average assessment processing time of
26.8 seconds - 11 AWS Lambda functions orchestrated
through Step Functions - Automated review of drafted loan
assessments using deterministic and
AI-assisted validation - Fully auditable rule-level outcomes with rationale, references, and status tracking
Background on the customer
Jacaranda Finance is an Australian consumer lender focused on delivering fast, reliable lending experiences while maintaining strong governance and compliance standards.
As application volumes continued to grow, maintaining quality assurance across drafted loan assessments became increasingly challenging. The business needed a scalable approach to review assessments consistently and efficiently while preserving transparency, auditability, and confidence in lending decisions
Challenge
Quality assurance of drafted loan assessments is a critical control within Jacaranda Finance’s lending process, helping ensure assessments align with internal credit policies, regulatory obligations, and operational standards before approval.
Historically, this review process relied heavily on manual assessment by QA specialists. While effective, the approach introduced operational bottlenecks as application volumes increased and created challenges around consistency, turnaround times, and scalability.
Key challenges included:
- High operational effort required to review each drafted assessment
- Variability in outcomes between reviewers
- Slower turnaround times during peak lending periods
- Difficulty scaling review capacity without increasing headcount
- Limited visibility into review rationale and decision consistency
- Growing compliance and governance expectations within a regulated environment
Jacaranda Finance needed a solution capable of automating assessment validation while preserving explainability, auditability, and confidence in review outcomes.
Stack Highlights
The Approach
Codex partnered with Jacaranda Finance to design and deliver a serverless AWS-native Quality Assurance Engine that automates the review of drafted loan assessments using a combination of deterministic business rules and AI-assisted validation.
Our approach included:
1. Working with lending subject matter experts to translate assessment review requirements into structured validation rules
2. Designing a scalable event-driven architecture using AWS Step Functions and Lambda
3. Building deterministic validation checks for objective policy, data quality, and calculation requirements
4. Leveraging Amazon Bedrock for contextual and judgement-based assessment reviews
5. Implementing automated ingestion and normalisation of assessment artefacts
6. Creating structured rule-level outputs with rationale, issue categorisation, and supporting references
7. Establishing audit logging, monitoring, and operational observability capabilities
8. Delivering infrastructure-as-code deployment pipelines using AWS CDK and GitLab CI/CD
This approach enabled consistent, repeatable, and transparent assessment reviews while maintaining flexibility for future rule enhancements and evolving lending requirements.
Technical Outputs
Codex delivered a production-ready automated QA platform that transformed how Jacaranda Finance validates drafted loan assessments.
Key outputs:
The platform provides a scalable and governed foundation for automated assessment review while reducing operational effort and increasing consistency.
QA Platform Highlights
Intelligent Loan Assessment Validation
Combines deterministic business rules with AI-assisted review capabilities to automate quality assurance across lending assessments.
Explainable and Auditable Outcomes
Every validation result includes rule-level rationale, references, issue tracking, and audit records to support governance and compliance requirements.
Rapid Processing at Scale
Average assessment review times reduced to just 26.8 seconds while maintaining high validation accuracy.
Serverless by Design
Built using Lambda, Step Functions, API Gateway, DynamoDB, and S3, allowing the platform to scale automatically while minimising operational overhead.
Designed for Regulated Financial Services
Governance, observability, auditability, and explainability were embedded from the outset to support lending operations and regulatory expectations.
Business and Commercial Outcomes
The automated QA platform delivered measurable operational improvements for Jacaranda Finance while creating a scalable foundation for continued growth and lending-volume expansion.
Key outcomes included:
- Achieved 96.26% validation accuracy on production validation datasets
- Reduced assessment review times to an average of 26.8 seconds
- Increased consistency and repeatability of QA outcomes
- Reduced reviewer effort through automated validation and issue identification
- Improved governance through auditable rule-level outputs and decision traceability
- Enabled lending teams to focus on exceptions and higher-value review activities
- Established a scalable architecture capable of supporting increasing application volumes without proportional increases in operational effort
- Created a foundation for future AI-enabled assessment and compliance capabilities
The platform now enables Jacaranda Finance to process assessments faster, improve quality assurance consistency, and scale lending operations with greater confidence while maintaining strong governance and compliance controls.
Talk to Us
We would love the opportunity to connect and understand more about the problems you are trying to solve.
Adrian Cambpell
Associate Partner, AI
Martin Campbell
Managing Partner
Get in touch to coordinate a meeting with one of our technical experts.
Australia: +61 7 3132 3002.
