What’s new here

AVA includes a pre-build planning layer before you write a line of code: a maturity assessment, operating model selection, use-case scoring against 25 weighted criteria, and a CFO-grade DCF model. These four interactive frameworks are persisted to DynamoDB and accessible from the Plan section of the platform. A fifth framework, Organization Design, translates the chosen operating model into a concrete org chart with roles, squads, and headcount ramp so that Plan hands Build a team, not just a strategy.

What it does

AVA is a web-based control plane (React frontend, FastAPI backend, Terraform infrastructure) that runs on ECS Fargate in your own AWS account. It organizes work across five pillars.

Plan gives you four interactive frameworks persisted in DynamoDB: maturity assessment, operating model picker, use-case prioritization, and business case builder, plus a fifth, organization design.

Build provides 35 FSI-specific multi-agent POC implementations (LangGraph and Strands Agents SDK variants), 9 reference implementations, 22 starter templates, and an App Factory that takes a plain-language brief and generates Strands agent code plus Terraform end-to-end.

Secure runs every Bedrock call through a LiteLLM proxy on ECS Fargate with virtual keys, budgets, rate limits, and Bedrock Guardrails presets tuned for financial services. A Cedar policy engine enforces per-tool access control.

Operate surfaces CloudWatch GenAI Observability, an embedded self-hosted Langfuse v3 instance, Bedrock Advanced Prompt Optimization, and an LLM-as-judge evaluation suite with promotion gates.

Govern covers 14 frameworks and 281 controls (SR 26-2, NIST AI RMF, EU AI Act, ISO 42001, and others), a risk register, model SBOM with differential-privacy tracking, shadow AI detection, and governance connectors to Azure, GCP, ServiceNow, Salesforce, and Copilot Studio. Of the 281 controls, 24 are assessed automatically today, and those checks only test that a control exists, not that it works.

Who it’s for

Financial services engineering teams at banks, insurers, and capital markets firms who are deploying agents on AWS and need to satisfy compliance and risk-management requirements, not just ship working code. Also useful for platform teams inside large enterprises who want a governed internal marketplace for AI resources rather than ad-hoc Bedrock deployments.

Try it

The README points to a platform architecture doc for full deploy instructions. You need an AWS account with Bedrock model access (Claude models enabled), AWS CLI 2.28.9 or later, Terraform 1.0 or later, Python 3.11 or later, Node.js 22 or later, and Docker with buildx. The control plane itself is deployed via Terraform, using the script the README lists as required:

cd platform/control_plane/infrastructure/scripts && ./deploy-full.sh

A quick-deploy path packages use cases from S3; a Git path uses CodeCommit for teams that want to fork before deploying. Only the Git path needs this optional step:

# Seed CodeCommit repos for the Git deployment path
cd platform/control_plane/infrastructure/scripts && ./seed-codecommit.sh init

After the control plane is up, every FSI Foundry use case and App Template deploys through the UI with one click via a CodeBuild + Step Functions pipeline.

How mature is it

67 stars, 22 forks, 8 contributors, 6 releases with the latest at v5.0 (2026-08-17). 29 commits in the last 90 days. 48 open issues and pull requests. Created 2026-02-06, last pushed 2026-09-28. Licensed Apache-2.0.