PitCrew verifies AI agent decisions with Automated Reasoning on AWS

To develop its solution, PitCrew met with financial services organizations across wealth management and broker-dealer operations. The same themes came up repeatedly: teams spending hours reconciling data across CRM, custodian, and billing systems; rebuilding the same reports from scratch every review cycle; and manually checking every document against policies and regulations. What’s more, compliance reviews, fee audits, and adviser compensation calculations rely on data spread across multiple systems, and every step requires human verification. Business rule interpretation, guardrail checks, and SOP analysis also depend on time-consuming and subjective expert judgment.

AI had the potential to help, but without a way to prove an AI output was correct, teams still had to verify every decision manually. Financial services organizations need automation that validates decisions with mathematical certainty. Until recently, applying this type of formal verification was an expensive and time-consuming process that relied on specialized scientists. Automated Reasoning makes it possible to encode natural language rules, such as financial services regulations, as mathematical logic that can prove the correctness of AI outputs.

PitCrew uses Amazon Bedrock to build generative AI applications and agents and Amazon Bedrock Guardrails to implement safeguards customized to application requirements. Using Automated Reasoning checks in Amazon Bedrock Guardrails, PitCrew converts regulations and business policies into logical statements that combine rules, citations, annotations, and variables into an Automated Reasoning policy. The reasoning engine evaluates each claim against these logical constraints, returning feedback that steers language models toward a definitive result instead of a confidence score. Each result includes a full explanation of how the conclusion was reached, giving compliance teams the transparency to trust and audit every decision.

PitCrew has prebuilt agents targeting front-, middle-, and back-office workflows. For example, the Form ADV review agent simplifies SEC registration. The agent takes a draft Form ADV as input, converts the form into claims, and runs them against an Automated Reasoning policy that encapsulates the Form ADV rules and instructions. Using the output, the agent flags the ADV form as ready to file or suggests corrections.

Anthropic’s Claude in Amazon Bedrock powers content extraction from complex documents including PDF forms, marketing HTML pages, emails, and invoices. PitCrew uses LangGraph to define agents that run within Amazon Bedrock AgentCore—a solution for production AI agents—using AgentCore Code Interpreter to build the tools each agent executes at runtime. PitCrew also uses AWS Lambda, a managed service to support faster development, improved performance, enhanced security, and cost efficiency. AWS Lambda orchestrates multistep processing and policy creation workflows, coordinating parallel extraction and managing the asynchronous policy build lifecycle. Amazon Simple Storage Service (Amazon S3) provides object storage for specification and proposal documents, and Amazon Relational Database Service (Amazon RDS)—a simple-to-manage relational database service—stores and tracks jobs and policy metadata. Customers can deploy the entire architecture in their own virtual private cloud, an important consideration for regulated industries where organizations manage the data plane on their own infrastructure.

Since launch, PitCrew has encoded 40 Automated Reasoning policies spanning multiple regulatory frameworks and business standards. Three production agents use combinations of these policies at runtime to verify marketing content and form submissions, reducing manual cross-referencing from 2 weeks to 30 minutes without sacrificing rigor. Every review provides compliance verification, full transparency into how conclusions are reached, and a complete, mathematically verifiable audit trail.

The impact extends across workflows. New client accounts that required 3–4 hours of retyping data across custodian, CRM, and compliance systems now open and verify in 10 minutes. Marketing and social media posts that sat in a 3-day review queue for SEC/FINRA compliance now clear in 30 seconds.

“LLMs give us the flexibility to understand any business process. Automated Reasoning checks on AWS give us the rigor to prove the output is correct,” says Rishi Kulkarni, CEO of PitCrew. “That’s what makes this work in a regulated industry.”

PitCrew’s architecture handles business workflows without proportional cost increases. With PitCrew agents automating the repetitive verification work, experts can focus on the higher-value work of judgment, reviewing edge cases, and customer-specific considerations.