Prediction Market Company

Platform Migration to AWS for a Prediction Market Company

Moved a prediction market platform from Google Cloud to a Terraform-managed, multi-account AWS estate built on EKS, RDS and MSK.

Prediction Market Company case study

Overview

The client is an onchain prediction market company whose platform lets users trade on the outcome of real-world events. Together with the client’s engineering team, DataMax built the AWS foundation for its production platform and migrated it from Google Cloud: a multi-account, Terraform-managed estate centred on Amazon EKS, RDS for PostgreSQL and Amazon MSK.

Challenge

The company’s platform had grown on Google Cloud: services on Cloud Run, stateful and chain workloads on VMs, PostgreSQL on Cloud SQL and a separately run staging cluster. The team wanted one Kubernetes-based platform across all environments, infrastructure that can be rebuilt from code, and no long-lived credentials on VMs. The difficulty was doing this for a live trading platform, with an order book to protect and a database where every row has to survive the move.

Approach

Production was never the first place anything ran. We built the AWS foundation first, then development and staging from the same Terraform code, and fixed every platform gap there. The cutover was scripted and rehearsed against real production data, so go-live was a known runbook rather than an improvisation.

Solution

The estate uses one AWS account per environment and workload type, with Terraform applied per account and a plan on every pull request. Applications run on Amazon EKS with dedicated node groups for the order book and the indexer, deployed from Helm charts. Data lives outside the cluster: RDS for PostgreSQL 16 in Multi-AZ, Amazon MSK for Kafka and ClickHouse on EC2. Databases were moved with scripted dump-and-restore jobs that run inside the VPC, and every run proves its result with exact per-table row counts. Nothing holds a long-lived credential: CI uses OIDC, pods use EKS Pod Identity, shell access goes through Session Manager and people sign in through IAM Identity Center. The blockchain nodes are their own stack in a separate account, peered to the application VPC. Prometheus, Grafana and CloudWatch cover metrics, alerts and logs, with Route 53 health checks watching from the outside.

Results

Production on AWS
The application platform runs on Amazon EKS, RDS and MSK instead of Cloud Run and Cloud SQL
Verified data move
Every production database checked table by table against its source before go-live
Repeatable environments
Development, staging and production built from the same Terraform code
Security by construction
No long-lived credentials, separate accounts per environment, network-isolated sequencer in production
Operational visibility
Dashboards, alert rules and outside-in uptime checks on the new estate

Ready to accelerate your AI journey?

Let's talk about your data and AI challenges. We'll help you build the right strategy and execute with speed.

Get in Touch