A payments monolith, one slice at a time.
A Fortune 500 fintech contracted a vendor to break its legacy payments platform into cloud-native microservices on Azure, built with Claude. A senior SNT engineer was embedded as the vendor's forward deployed engineer, alongside the company's director of software engineering. Our engineer architected the platform and delivered the first pilot with the vendor's offshore team, all new to Claude.
What the work covered
A plan for the whole platform
Requirements, system design and a technology stack for moving a 53-project .NET Framework monolith, out of support in 2027, to .NET 8 microservices on Kubernetes in two regions, one database per business domain.
One payment first
A pilot called Hello Payment: one card payment, through one gateway, for one existing client, built with Claude and running end to end. Small enough to finish, complete enough to fix the standards every later service follows.
Standards before scale
How services call each other, how they reach their data with no passwords in code, how every request is traced, and how each payment provider sits behind its own adapter, all settled on the pilot before the other 60 to 70 APIs move.
The numbers
The old system’s figures are its baseline and the team figures are actuals. Every figure for the new platform is a design target, not yet measured.
- 53
- projects in the old monolith, rebuilt for any change
- None
- of the offshore team had used Claude or agentic engineering before
- 6 + 2 weeks
- ramp-up, then the pilot built with Claude
- 47 → < 3
- database calls per payment (design target)
- 17 → 500
- payments per second, sustained (design target)
- ~$1.2M → ~$0.8M
- yearly running cost (design target)
How the work was done
- 01
Baseline the old system
Measure what holds it back: build times, database calls per payment, throughput, running cost and the weeks it takes a new developer to find their way around.
- 02
Design the target
Requirements, system design and stack written down and agreed with engineering leadership before any code.
- 03
Pick one vertical slice
One payment flow, end to end, so the team learns every layer without getting stuck in cross-domain dependencies.
- 04
Spec each feature
Each piece of the pilot got a goal, a boundary, an integration pattern and a definition of done the offshore team could build and test against.
- 05
Then scale out
Four horizons: split the data, put the new front door in place, move the services, then move merchants across in small groups with checks and a way back.
Built for payments. Card data stays as tokens, so the core services stay out of card-data scope. Providers are isolated so one failure cannot take down the rest, services reach data with managed identities instead of passwords, every payment carries one tracking ID from the front door to the ledger, and existing merchant integrations keep working unchanged.
See it, then talk to us
This was our engineer’s work inside a vendor team, not an SNT Agents build. The pilot was built with Claude by the vendor’s offshore team. Ask us how the same path would work for your platform.