Building a fraud detection console with MongoDB and Mastra

​Fraud investigations are slow and hard to defend. Evidence is scattered across logs and data stores, reviews are inconsistent, and auditors need a reproducible trail to follow. That makes fraud a natural fit for agents, with a human in the loop.

​Marshal is a fraud investigation console built with Mastra and MongoDB Atlas. It uses multiple agents and real-time data streaming to automate anti-money laundering (AML) and fraud workflows. Every decision is human reviewed, auditable, and reproducible.

​In this workshop, we'll walk through Marshal's architecture and how it uses Mastra's durable workflows to run each investigation.

​You'll learn:

• ​Evidence modeling in MongoDB Atlas across vectors, text, and graph

• ​Retrieval and ranking that combines vector, text, and graph signals

• ​Durable workflows with suspend/resume, snapshots, and human approval gates

• ​Observability and PII controls, including tracing, redaction, selective escalation, and production tuning

​Followed by live Q&A

​Who is this workshop for?

​Engineers building agents for fraud, AML, or any workflow where every decision needs an audit trail. No prior Mastra experience needed.

​Hosted by

• ​Brandon Barros, Founding Product Advocate, Mastra

• ​Nina Lopatina, Staff Developer Advocate, MongoDB

​Recording, reference architecture, and runnable code samples will be available to everyone who registers.