Loading...
Back to case studies
Client project
AI-powered mortgage broker assistant for eThink Solutions
A production-grade RAG and graph retrieval system for the Australian mortgage industry, helping brokers query operational documents with grounded, source-aware responses.
Workflow evidence
Screenshots from the systems and workflows.
These visuals come from the project portfolio and show the workflow, dashboard or training assets behind the case study.
Problem
The operational problem behind the work.
Mortgage brokers work across dense policy, process and lender documentation. The client needed a way to reduce manual lookup time while keeping answers traceable to source material and suitable for a compliance-sensitive workflow.
Approach
How the system was shaped.
01
Ingested and structured more than 150 loan and operational documents.
02
Designed a retrieval workflow that combines document chunks, relationships and natural-language querying.
03
Used graph modelling to represent relationships between banks, processes and compliance requirements.
04
Built orchestration around AWS Bedrock, ArangoDB and n8n so the workflow could move beyond a proof of concept.
Architecture
Components that made the work production-minded.
Document ingestion and chunking pipeline
ArangoDB graph model for document relationships
AWS Bedrock model layer for language understanding and generation
OpenSearch and retrieval components for source-aware responses
n8n orchestration connecting ingestion, retrieval and response steps
API Gateway and Lambda components for production access patterns
Outcomes
What changed or became possible.
150+ loan documents processed
95%+ accuracy reported on loan document understanding
80% reduction in manual lookup time
Natural-language query interface for mortgage knowledge retrieval
Responsible AI
Why governance is part of the implementation.
Grounded responses designed around source material rather than unsupported model output.
Architecture supports traceability and review for compliance-sensitive mortgage workflows.
System boundaries help keep the assistant as a broker support tool, not an unchecked decision-maker.
Stack
Tools and platforms used.
AWS BedrockArangoDBn8nAWS LambdaOpenSearchAPI GatewayRAG
Similar challenge?
Start by assessing the workflow, data and risk.
The AI Readiness Audit is the cleanest first step when you need to decide what to automate, what to avoid, and what should be governed before a production build.