An agentic financial reporting system, built on AWS

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

The challenge had three focus areas, each targeting a different part of the reporting journey.

At the centre of the solution is a multi-agent system built with Strands and orchestrated on Amazon Bedrock AgentCore. Rather than a single model attempting everything, the work is divided across specialised agents: a Report Generation Agent that produces agreed packs and narratives, a Report Comparison Agent that computes deltas and summarises drivers across periods and entities, a Natural Language Query Agent that classifies intent and plans queries, a Data Quality and Reconciliation Agent that flags missing or inconsistent data, and a General Inquiry Agent that handles routing, report discovery and permissions guidance.

The engagement delivered the agentic reporting system defined in the project success criteria, with the multi-agent architecture, natural language workflows and reporting integrations implemented and validated against the client's acceptance process. Acceptance was based on demonstrating the report generation, comparison and natural language query workflows end to end, and confirming that the AWS environment is suitable for the continued development of the client's service.

• Report generation on demand — A Report Generation Agent produces agreed packs, summaries and variance narratives from underlying datasets, replacing the long manual workflow with a plain language request.
• Natural language querying and comparison — Users can ask questions in plain English and compare reports across periods and entities, with deltas, anomalies and drivers surfaced automatically.
• Explainable, attributed outputs — Results are returned in a consistent structure of tables, narrative and key metrics, each traceable back to its source data to reduce manual quality-assurance effort.
• Memory that improves with use — Amazon Bedrock AgentCore Memory retains session context and extracts long-term insights such as preferred metrics and recurring prompts.
• Handover — The solution was provisioned via infrastructure as code with access controls and audit logging, and delivered with technical documentation, workshops and recorded knowledge-transfer sessions.

With a validated multi-agent foundation in place and hosted in the client's own AWS environment, the platform is positioned to extend to further reporting use cases and to grow alongside whatever the client takes on next.