Case Study

Weeks → Minutes

Governed AI Agent Builder on AWS for Real-Time GRC & Audit Readiness

How Cogninest AI built an enterprise-grade agent platform on Amazon EKS and Bedrock with fail-closed audit ledgers and HashiCorp Vault security.

Before
Compliance Policy Drafting2 - 4 weeks manual writing
AWS Cloud Security ChecksManual console reviews
Custom Workflow CreationEngineering sprint backlog
Audit Trail IntegrityAd-hoc spreadsheets
After
Compliance Policy DraftingDrafted in minutes
AWS Cloud Security ChecksAutomated Boto3 discovery
Custom Workflow CreationZero-code in plain English
Audit Trail IntegrityFail-closed signed ledger
01The Challenge

Repetitive compliance audits and manual cloud reviews stalled growth

Security certification and audit preparation (SOC 2, ISO 27001, NIST CSF) involve immense repeated manual effort: writing policies that match control frameworks, collecting evidence from cloud infrastructure, and maintaining audit logs. Small and mid-sized businesses struggled with generic document templates that failed auditor scrutiny, while enterprise clients required air-tight multi-tenant isolation, tamper-evident action recording, and zero exposure of sensitive cloud credentials to model prompts.

  • Manual audit preparation consumed hundreds of hours across every certification cycle.
  • Generic policy templates failed to match actual cloud configuration realities.
  • Customer cloud credentials risked exposure if placed in model prompts or standard DBs.
  • Auditors demanded tamper-evident, verifiable proof for all automated actions.
02What We Built

Governed Agent Platform with Isolated Tool Execution & Vault Security

Cogninest AI architected and built a multi-tenant Generative AI agent platform on AWS using Amazon EKS, Amazon Bedrock, and OpenTofu infrastructure-as-code. The system cleanly decouples agent reasoning from tool execution, ensuring customer credentials never enter model prompts.

Step 1: Conversational Workflow & Policy Builder

Built an intuitive conversational builder where users describe compliance goals in plain English, creating editable LangGraph workflow plans powered by Claude Opus.

Step 2: Isolated Tool-Executor & Vault Secrets

Engineered an isolated tool-executor service using HashiCorp Vault ephemeral credentials; customer AWS keys are never stored in application databases or model prompts.

Step 3: Framework-Grounded RAG with Qdrant

Deployed self-hosted Qdrant on Amazon EKS with Amazon EBS and Amazon Titan Text Embeddings V2, enabling tenant-isolated retrieval across SOC 2, ISO 27001, and company docs.

Step 4: Tamper-Evident Fail-Closed Audit Ledger

Implemented an immutable, hash-chained audit ledger in a dedicated PostgreSQL instance with Vault Transit Merkle signing—blocking any tool action if audit logging fails.

Amazon BedrockAnthropic Claude (Opus, Sonnet, Haiku)Amazon Titan Embeddings V2Amazon EKSQdrant Vector DBHashiCorp VaultPostgreSQL (Row-Level Security)LangGraphFastAPI & Next.jsAmazon ElastiCacheLangfuse & OpenTofu

"Users can now create outcome-driven, governed AI agents and multi-step workflows to automate GRC use cases, going from "what do we need for SOC 2?" to drafted policies and live AWS remediation in minutes without writing code."

Head of Product

Governance, Risk & Compliance Technology Platform

03The Outcome

Continuous Compliance and Instant Audit Readiness

The platform converted compliance from an annual, dread-inducing scramble into an automated, continuous, and verifiable background process.

Minutes

to draft framework-mapped policies

Zero-Code

custom compliance workflow creation

100%

fail-closed tamper-evident ledger coverage

Multi-Tenant

row-level security and tenant-isolated RAG

Automated

AWS cloud configuration discovery

Want to Build Governed, Enterprise-Ready AI Agents?

Discuss your compliance, security, and agentic AI requirements with the Cogninest AI engineering team.