Building a Deploy-Anywhere AI Platform for MetisJean on AWS

Introduction

MetisJean is redefining enterprise intelligence with its sovereign AI platform. The company’s mission is to give organizations genuine ownership over their data and the AI models that reason over it. Unlike traditional SaaS AI tools that process data externally, MetisJean’s platform is designed to be deployed directly within a customer’s own infrastructure. This sovereign-first approach eliminates vendor lock-in and ensures that sensitive corporate knowledge never leaves the customer’s control. To deliver on this promise the company needed a foundational platform that was both powerful enough to support a sophisticated AI product and portable enough to run anywhere.

The Challenge

As an early-stage software provider, MetisJean faced a dual challenge. First, they needed to establish a secure, scalable and automated cloud foundation on AWS to support their own rapid product development. Second, their core product had to be architected for portability, capable of being deployed seamlessly into diverse customer environments including other cloud providers and on-premises data centers. The architecture needed to be repeatable, easy to support and scalable.

This required a sophisticated approach to DevOps and Data Engineering from day one. The infrastructure, deployment pipelines and operational tooling had to be built for consistency and automation ensuring that creating a new customer environment would be a fast, reliable and low-risk operation.

Our Solution

GOStack partnered with MetisJean to design and build a new platform on AWS that met their specific needs for a portable, secure and fast platform. The solution covered everything from the DevOps lifecycle to the complete AI and data stack creating a production environment that now serves as the blueprint for all customer deployments.

A Greenfield DevOps Foundation with Terraform and AWS Control Tower

GOStack established a secure AWS Organization with strong governance using AWS Control Tower. This provided a landing zone with pre-configured settings for security, networking and identity management through AWS IAM Identity Center. The entire infrastructure from VPCs to EKS clusters was defined as code using Terraform. This enforced a strict Infrastructure-as-Code (IaC) model, meaning no manual changes were made in the AWS console. To automate the creation of new customer environments GOStack built a custom Account Factory for Terraform using AWS Step Functions and Lambda to orchestrate the provisioning of new AWS accounts with all the necessary guardrails and configurations applied automatically.

Portable Kubernetes-Native Architecture with Amazon EKS and GitOps

Amazon EKS was chosen as the compute backbone over other services like Amazon ECS because its industry-standard Kubernetes API ensures the application layer remains portable across any compliant Kubernetes cluster. This was critical for MetisJean’s deploy-anywhere strategy.

A complete GitOps workflow was implemented using ArgoCD for continuous delivery and GitHub Actions for continuous integration. When code is merged in GitHub a pipeline automatically builds and scans the container image pushing it to Amazon ECR. ArgoCD detects the new image tag in the Git configuration repository and automatically synchronizes the deployment to the target EKS cluster. This hands-off automation dramatically reduces deployment time and eliminates manual errors.

To optimize cost and performance the EKS clusters use Karpenter to dynamically provision nodes based on actual workload demand. Karpenter automatically selects from a mix of AWS Graviton, memory-optimized and compute-optimized instances ensuring compute capacity is always proportional to the workload at any given time.

A Sovereign AI and Data Platform with Amazon Bedrock and OpenSearch

The platform’s intelligence is powered by a Retrieval-Augmented Generation (RAG) architecture. GOStack integrated Amazon Bedrock to provide secure, private access to a range of foundation models without the operational overhead of self-hosting. This allows MetisJean to leverage powerful AI capabilities while ensuring customer data is never used for model training.

The data layer is built on Amazon S3 and Amazon OpenSearch Service. Data ingestion pipelines store raw documents and processed data in an Amazon S3 data lake. The data is managed using the Apache Iceberg open table format which provides robust schema evolution and efficient data management for large-scale analytics. This allows for serverless ad-hoc querying of the data lake using Amazon Athena offering powerful analytics capabilities without the need to manage underlying infrastructure. Amazon OpenSearch Service is used to index enterprise content and vector embeddings to support the RAG architecture’s semantic and vector search capabilities. For the initial AWS deployment using the managed OpenSearch service reduced operational complexity and provided built-in high availability and durability compared to self-managed search clusters. This data architecture ensures that for deployments outside of AWS these managed components can be swapped with self-hosted equivalents without changes to the application layer.

Results and Benefits

The partnership with GOStack enabled MetisJean to build a sophisticated, enterprise-ready platform.

• Enabled Real-Time Executive Insights: The time required for executives to get answers to complex business questions dropped from hours or days to under 5 seconds.

• Portable & Sovereign-Ready: The EKS-based architecture provides a consistent, portable application layer that can be deployed in any customer environment.

• Fully Automated & Self-Service: Developers are empowered to build, deploy, and manage services independently, accelerating feature velocity.

The Infrastructure for true AI Sovereignty

GOStack is our go-to partner. They gave us the infrastructure to deliver on our promise. We can now deploy in any customer environment, whether cloud or on-premises. Sovereignty is what makes us competitive.

Alexander Haywood

CEO & Founder

MetisJean