Case study – Delivering a governed AI gateway at enterprise scale

The challenge: securing and governing enterprise-wide AI usage

A large enterprise organisation needed to enable safe, governed access to generative AI across dozens of separate teams and hundreds of users. Without a central platform, each team would procure its own LLM access, creating ungoverned adoption with no data protection, no spend visibility, no consistent safety controls, and no data residency guarantees. The compounding risk was sensitive data reaching models unprotected, uncontrolled spend, and fragmented adoption that made policy enforcement and compliance impossible.

A centralised, secure gateway for enterprise AI adoption

Version 1 designed and deployed a production-grade, centralised AI gateway built on Amazon Bedrock as the core inference layer, hosted entirely within the customer’s own AWS environment. A proxy layer running on Amazon ECS Fargate handles model routing, per-team spend tracking, and budget enforcement, while Amazon Cognito, AWS WAF, Amazon Aurora, and Amazon ElastiCache provide authentication, perimeter security, and state management. Bedrock Guardrails are applied to every inference request, enforcing content filtering, prompt-attack detection, and PII anonymisation gateway-wide so no client can bypass them. The entire estate is provisioned as code for repeatable deployment.

From uncontrolled AI to governed, enterprise-wide adoption

The platform moved the organisation from zero governed AI access to active adoption across 50+ teams and hundreds of users, with responsible AI controls applied to 100% of inference requests. Centralised budget enforcement and per-team cost visibility replaced uncontrolled spend with a transparent, accountable model, and shared infrastructure amortised across all teams kept costs well below the alternative of each team running its own platform.