Internal enterprise AI platform

Internal Enterprise AI Catalog Modernization

Deployed internal

Overview

Modernized the AI layer of a deployed internal catalog on AWS, moved model use to Amazon Nova Pro, and strengthened guardrails, access controls, and Terraform-managed infrastructure.

Context / Problem

A deployed internal catalog needed its model layer refreshed while preserving enterprise access patterns and the operational expectations of an existing application. The change had to improve model use and protective controls without exposing internal identifiers or weakening the platform boundary.

My Role

I owned the AI-layer modernization and contributed hands-on implementation, infrastructure changes, security controls, validation, and the technical decisions needed to keep the deployed application coherent.

What I Did

Migrated application model use to Amazon Nova Pro through Amazon Bedrock and adapted the surrounding integration.
Implemented guardrails and access-control measures around the AI interaction path.
Maintained the relevant AWS infrastructure through Terraform and validated the deployed internal behavior.

Architecture

Authenticated users pass through the enterprise identity boundary into the web and API layer. The AI application invokes Amazon Bedrock and Nova, while managed data and search services support catalog behavior. Terraform defines the relevant cloud resources and the security boundary surrounds the application and model path.

Deployed AI catalog request pathA public-safe reconstruction of the authenticated request path, model invocation, supporting services, and infrastructure boundary.

UserImplemented
Enterprise SSOImplemented
Web / APIImplemented
AI applicationImplemented
Bedrock / NovaImplemented
Data / searchImplemented
Terraform controlsImplemented

Guardrails and access controls apply across the application and model path.

Key Decisions / Trade-offs

Modernize within the deployed platform: Changing the model layer in place preserved the application's enterprise context while allowing a bounded modernization of AI behavior.
Treat controls as part of the request path: Guardrails and access controls were integrated with the AI flow instead of being described as an external assurance activity.

Implementation Scope

The model integration, protective controls, and supporting Terraform changes were implemented for an internally deployed application. The work concerned the AI layer and its immediate platform dependencies rather than a wholesale rebuild of every application component.

Safety / Security / Governance

Normal enterprise authentication remained the entry boundary. Guardrails, access control, and infrastructure-as-code review supported controlled operation. Public reconstruction omits internal domains, account data, repository names, identifiers, and private operational detail.

Scope Boundary

This case describes implemented modernization within a deployed internal application. It does not claim ownership of the whole product, disclose private platform details, or attribute unsupported delivery metrics. Supporting API, data, and delivery-pipeline details are kept generic where the public evidence boundary is narrower.