Mintzy · 30 Oct – 3 Nov 2026
Moving a live trading platform to a safer cloud in five days
Mintzy's trading platform ran on servers exposed to the internet in AWS Virginia. We moved it to a private, self-healing setup in AWS Mumbai, closer to Indian brokers, with keyless deploys and every change approved first.
- from first review to trading again
- 5daysfrom first review to trading again
- cloud resources written as code
- 333cloud resources written as code
- database records moved, restore tested
- 7.3L+database records moved, restore tested
- passwords or keys left in GitHub
- 0passwords or keys left in GitHub
Where it started
- Eight apps on servers with public IPs, deployed by logging in over SSH
- No backups of servers or data
- No audit trail or alerts when something went wrong
- Passwords stored inside code repositories
- Services talking to each other in plain HTTP across the internet
Where it is now
- No public servers. Only the load balancer and fixed outbound IPs face the internet
- Self-healing: a broken server is replaced from its image in minutes
- Daily and weekly backups, copied to a second region for disaster recovery
- Audit logs, threat detection and alarms, with every server session recorded
- Secrets in AWS Secrets Manager, none in code
- Fixed IPs per broker customer, so brokers can allowlist orders
What we built
Private network · Mumbai · 2 availability zones
No public IPs, no SSH, encrypted disks. Each service self-heals.
- backend ×2
- gateway
- signal
- kelly
- kelly-async
- bear-street
- trader-hub
- models
How we did it
- 1
Review
A read-only security and cost review of every app, server and repository.
- 2
Design
Two availability zones, private subnets, internal load balancer and Multi-AZ Redis behind a firewall.
- 3
Build as code
333 resources in Terraform, checked automatically: no public IPs, no open SSH, everything encrypted.
- 4
Safety switches
Both regions shared one database, so apps, schedulers and live orders stayed off until each was approved.
- 5
Cut over
Final snapshots, a verified database restore, then the switch to Mumbai with checksummed state.
- 6
Keyless deploys
GitHub signs in to AWS with short-lived OIDC credentials. Failed deploys roll back on their own.
What we learned
Self-healing cuts both ways
Anything on a server's disk that isn't in its image is lost on replacement, so the tooling has to make the safe path the easy one.
Inventory the data outside the database
Session addresses, token files and trade logs all needed moving too. We found them before the move, not after.
Test the pipeline itself
A credential scan matched its own pattern and would have failed every run. We caught it in testing.
- AWS
- Terraform
- GitHub Actions
- OIDC
- SSM
- CloudWatch
- GuardDuty
- AWS Backup
- Redis
- WAF