How DXC Technology built a multi-model AI substrate on Amazon Bedrock
DXC Technology built a governed multi‑model AI substrate on Amazon Bedrock that unifies access to Anthropic Claude, Amazon Nova, Meta Llama and other open‑weight families via a single API, while enforcing Bedrock Guardrails for content filtering, PII redaction, and denied‑topic policies. System‑level controls such as AWS PrivateLink keep inference traffic inside the customer VPC and serverless scaling removes the need for self‑managed GPU infrastructure. Since its GA in March 2026, the DXC OASIS platform has cut resolution time by 95 % across 56 measured use cases, demonstrating enterprise‑grade resilience and governance. Engineers can leverage this substrate to safely orchestrate AI agents at scale without building custom model‑access layers.
⚡ Key Takeaways
- Resolution time reduced by 95 % across 56 use cases after deploying DXC OASIS.
- Bedrock Guardrails provide uniform content filtering, PII redaction, and denied‑topic enforcement across all models.
- Serverless scaling eliminates the need for self‑managed GPU clusters, reducing infrastructure failure points.
- Integration uses a single consistent Amazon Bedrock API to invoke multiple foundation models.
- Customer prompts and completions remain within the AWS data boundary and are not used to train the underlying models.
The substrate lets engineers deploy AI agents in regulated environments (banks, hospitals, government) with built‑in security, governance, and model‑agnostic flexibility, accelerating time‑to‑value while mitigating the risk of uncontrolled autonomous actions.
✅ Practical Steps
- Invoke foundation models through the Amazon Bedrock API, selecting from Claude, Nova, Llama, etc., to leverage the unified interface.
- Enable Bedrock Guardrails for content filtering, PII redaction, and denied‑topic policies to enforce enterprise compliance.
- Configure AWS PrivateLink for the Bedrock endpoint to keep all inference traffic within the VPC and avoid external exposure.
Want the full story? Read the original article.
Read on AWS ML Blog ↗