About Acumen Data
Acumen Data is an Australian data management company specialising in accounts payable processing and procurement analytics. Their CAAPS(R) (Complete Accounts Payable Solution) platform manages purchase orders, invoices, vendor information, and freight charges for enterprise clients, processing hundreds of thousands of transactions across 25+ database tables.
Customer Challenge
Acumen Data faced a significant barrier in making their procurement data accessible to business users. The CAAPS(R) database contained valuable insights about purchase patterns, cost trends, and vendor performance, but extracting this information required specialised SQL knowledge and deep understanding of complex table relationships. This created a dependency on technical staff for routine data queries, slowing decision-making and limiting the ability of procurement teams to perform ad-hoc analysis.
The company needed a solution that would enable non-technical users to query procurement data using natural language, asking questions like “What should we pay for product X?” or “Show me freight costs over the last 6 months”, without requiring SQL expertise or knowledge of the underlying database schema.
Partner Solution
DNX Solutions, an AWS Premier Consulting Partner, delivered a production-ready AI-powered procurement analytics system using AWS Bedrock Agents. The solution enables natural language queries over the CAAPS database through an intelligent agent architecture that interprets user questions and orchestrates the appropriate database operations.
The core system features a two-layer Lambda architecture: a Toolbox Lambda providing seven specialised action group functions (product code lookup, quantity analysis, cost statistics, freight analysis, and custom SQL execution), and an RDS Lambda handling secure database connectivity via pyodbc with ODBC Driver 17 for SQL Server. The agent, powered by Amazon Nova Pro, uses comprehensive prompt engineering developed through six validation cycles to achieve 90% accuracy on standardised test queries.
DNX implemented a Streamlit web interface for user interaction, integrated with Langfuse for real-time observability of agent traces, tool usage, and performance metrics. The Streamlit application uses the AWS Bedrock Agent Runtime API to invoke the agent with streaming responses, displaying the agent’s reasoning process and tool invocations in real-time. The entire solution is containerised using Docker Compose for simplified deployment and management.
Results and Benefits
The implementation achieved all target metrics through systematic iterative development across six validation cycles:
- Response Time: 24.98 seconds average (target: less than 30 seconds)
- Accuracy: 90% on standardised test queries (target: greater than 70%)
- Cost: $0.12 per interaction (74% reduction from initial prototype)
The solution will transform how CAAPS(R) users access procurement information. Business users could now ask questions in plain English, such as “What quantity of LPG did we purchase in the last three months?” or “Has the price for product 000 0100-030 increased or decreased?”, and receive comprehensive responses including statistical summaries, sample invoices, and direct document links.
The six-validation-cycle development methodology proved that production-ready AI systems require balanced optimisation across accuracy, speed, and cost. By combining Amazon Nova Pro’s efficiency with specialised tools and advanced prompt engineering, DNX delivered a system that exceeds requirements while remaining commercially viable.
About the Partner
DNX is a systems integrator and trusted advisor for business transformation, focused on democratising modernisation. For us, modernisation is not just about upgrading technology. It is about enabling organisations to adopt modern platforms, architectures and ways of working that drive measurable business outcomes. This includes cloud-native technologies, security and compliance by design, data-driven decision-making, and governance models that support scale, agility and innovation. By democratising modernisation, DNX ensures these capabilities are accessible, practical and aligned to each organisation’s maturity—empowering teams across the business, not just IT, to modernise with confidence and purpose.
Architecture Diagram
AI-powered procurement analytics leveraging AWS Bedrock Agents, Lambda, and RDS SQL Server.
- User Interface: Web interface (Streamlit) for natural language queries
- GenAI Service: AWS Bedrock Agent with Amazon Nova Pro for reasoning and orchestration
- Action Group: Toolbox Lambda with 7 specialised functions for procurement analytics
- Database Layer: RDS Lambda with pyodbc for secure SQL Server connectivity
- Database: Amazon RDS (SQL Server) hosting CAAPS procurement data
- Monitoring: Langfuse for agent observability and performance tracking
AWS Services Used
| Service | Purpose |
|---|---|
| Amazon Bedrock | Foundation model hosting (Nova Pro) |
| AWS Bedrock Agents | Agent orchestration and tool management |
| AWS Lambda | Serverless compute for action group and database functions |
| Amazon RDS | SQL Server database hosting CAAPS(R) |
| Amazon CloudWatch | Logging and observability |
| AWS VPC | Network security and Lambda-RDS isolation |
Plan Your Next Move with Confidence
Ready to align your technology with your business growth strategy? Talk to DNX about modernising your platform for scalability, resilience, and faster time-to-market.
