Ninth Wave has deployed an AI-powered onboarding assistant that cuts API mapping and analysis time by 95% for financial institutions joining the open finance network. The system, called Compass, processes weeks of specialist work in minutes, addressing a persistent bottleneck where banks must validate and map their unique APIs to the Financial Data Exchange standard before connecting to aggregators like Plaid, Finicity, and MX.
How the multi-agent system works
Compass runs on Amazon Bedrock AgentCore and uses a multi-agent architecture where each specialist handles a single task. The orchestrator classifies user intent once and routes the request to one of seven agents: search, documentation Q&A, document classification, field mapping, analysis, interactive workflows, or readiness analysis. Each agent operates within a bounded context window, which prevents prompt dilution and keeps outputs predictable.
Before any agent is invoked, the application assembles that specific bank's context from Amazon OpenSearch Service and Amazon S3. This tenant-scoped grounding means one bank's API documentation and configuration data never enters another bank's session, even though all tenants share the same model infrastructure. The readiness score itself is computed deterministically from field-mapping coverage in OpenSearch, not estimated by a probabilistic model, satisfying audit requirements.
Model selection and security design
The team matched model capability to task complexity rather than routing everything through a single large language model. Lightweight models handle high-volume tasks, while higher-reasoning models handle mapping, analysis, and interactive Q&A. The AI workloads run in a dedicated AWS account, separated from application workloads by a cross-account IAM role for blast-radius containment.
Security hardening was a dedicated implementation phase. The system enforces OAuth2/OIDC with multi-factor authentication, TLS 1.2+ at the edge and TLS 1.3 internally, customer-managed KMS keys for data at rest, and per-tenant data isolation across OpenSearch indices and S3 prefixes. Application-layer behavioral and safety constraints are applied per agent, giving the team the ability to tune scope and output boundaries for one specialist without affecting others.
Production results
The system tracks invocation volume, token usage, latency, and cost by agent through CloudWatch metrics fed into Grafana dashboards. The CI/CD pipeline runs through GitHub Actions with a circuit-breaker rollback that automatically reverts deployments failing health checks - a safeguard when prompt or grounding changes could affect agent accuracy before metrics surface it.
"AI fundamentally changes how financial institutions build, integrate, and innovate, but it requires trust in the underlying infrastructure," said George Anderson, Founder and CEO of Ninth Wave. "Compass represents our vision for bringing intelligence, governance, and automation together - enabling banks to onboard faster, reduce operational complexity, and build the trusted foundation required for the next generation of AI-powered financial services."
Compass launched in beta on March 1, 2026, with general availability on June 1, 2026. Each bank receives a custom-branded developer portal to invite aggregators and fintech partners to self-serve.
Why this matters for finance professionals
Open finance connectivity has been a manual, multi-week engineering slog. A 95% reduction in API mapping time changes the economics of bank-fintech partnerships directly. For finance leaders evaluating AI investments, the architecture decisions here - per-task model selection, deterministic scoring for compliance, and tenant-scoped grounding - offer a concrete pattern for deploying governed AI in regulated environments. The AI Learning Path for CFOs covers similar strategic decisions around automation and compliance in financial services.
Your membership also unlocks:
