LangGraph and Strands power market surveillance AI on AgentCore
Multi-agent orchestration meets production-ready compliance with human-in-the-loop recovery.
Financial market surveillance demands multi-agent AI systems that coordinate specialized tasks—analyzing trades, investigating suspicious patterns, and generating reports—while maintaining strict compliance and reliability. Single-agent approaches fail at such complex, stateful workflows. This solution combines LangGraph for macro-level orchestration (directed graphs with shared state, checkpoints, and human-in-the-loop) with Strands for agent reasoning (model-agnostic, tool integration, and session management). Amazon Bedrock AgentCore simplifies production hosting on AWS.
The implementation separates data discovery from retrieval using tools like get_report_list and get_report_schema to avoid hallucinations and injection attacks. Strands agents run within LangGraph nodes, using a reasoning loop that evaluates tool outputs dynamically. The example uses Claude Sonnet 4.6 via Bedrock with adaptive thinking. LangGraph’s persistence layer enables checkpoint-based recovery from failures, making the system robust for real-world trading floors. The entire codebase is available as a GitHub repository, demonstrating a reusable pattern for enterprise-grade agentic systems.
- LangGraph manages multi-agent state and execution with checkpoint-based recovery and human-in-the-loop support.
- Strands Agent provides model-agnostic reasoning with tool integration, session management, and caching (e.g., Claude Sonnet 4.6 with adaptive thinking).
- Separation of data discovery from retrieval (get_report_list, get_report_schema, run_report) reduces hallucination and injection risks in financial surveillance.
Why It Matters
Enables enterprises to deploy reliable, observable multi-agent AI for compliance and production workloads.