nearcore ITBook a call
CLOUD · DEVOPSFINTECH · DACH2025—2026

Replatformed a 12-year-old monolith without freezing the roadmap.

Atlas Bank had a Kubernetes mandate from the board and a feature backlog that couldn't wait. We embedded three SREs and a platform PM for nine months, alongside their existing team.

DEPLOY FREQUENCY
41×/wk
from 3×/wk
BUILD TIME
6 min
from 38 min
MTTR
22 min
from 4h 12m
ROADMAP DELIVERED
94%
on plan
CLIENT
Atlas Bank AG · Frankfurt
ENGAGEMENT
Embedded pod · 9 months
TEAM
3 SREs, 1 Platform PM
STACK
AWS, EKS, Terraform, ArgoCD
/ THE BRIEF

Two clocks, ticking against each other.

The CTO's mandate was clear and impossible at the same time: move the core servicing platform to Kubernetes by Q3, without slowing down the trading-product roadmap. Atlas had tried this once before with a Big Four firm — six months in, no production workload, three abandoned Terraform modules.

The internal team was good. They were also fully booked on the regulatory backlog from the 2024 audit. Adding "rebuild the platform" to their plate would have meant choosing which deadline to miss.

We weren't short on smart people. We were short on smart people with weeks free this quarter.
/ THE WORK

Nine months, three phases.

01 · WEEK 1–6

Land, read, prove the path

Two of our SREs were in Atlas' Slack on a Monday. By Wednesday we'd read every existing Terraform module, every Confluence runbook (and a few that weren't in Confluence), and the audit report.

The first PR shipped Friday of week one — a CI fix for the build that had been failing intermittently for four months. Trust came from that, not from a kickoff deck.

By week six we had EKS in three environments, a working ArgoCD pipeline, and one non-critical service in production. The CTO had a deploy demo for the board.

02 · MONTH 2–6

Lift & reshape, two services at a time

We didn't do a big-bang migration. Each fortnight we picked two services from the monolith, rewrote the parts that needed cloud-native (autoscaling, eventing, managed Postgres), left the rest, and shipped them behind feature flags.

Atlas' own engineers paired with us on every migration. The deal was: by service four, an Atlas engineer leads the migration and one of ours reviews. By service eight, we're reviewing only at the architecture level.

The trading-product roadmap kept moving. Three of our engineers were on infrastructure; the fourth was a senior backend who unblocked the team's biggest spec while we were there.

03 · MONTH 7–9

Hand it back, then leave

Last six weeks: zero new infrastructure work, all knowledge transfer. We rewrote the runbooks against the platform we'd built, ran two game-days, and put their on-call rotation through every failure mode we could think of.

The handoff document was 38 pages. The handoff meeting was 90 minutes. We rolled off in October. Atlas owns the platform.

"
Two senior SREs landed in our Slack on a Tuesday. By Friday they'd cut our deploy time in half and rewritten the runbook nobody on our side had ever finished.
EW
Elena Weber · VP Engineering
ATLAS BANK AG · FRANKFURT
/ WHAT WE'D DO AGAIN

Three things that made it work.

  • Ship in week one, even if it's small. Trust compounds from real PRs, not from architecture diagrams.
  • Pair every migration with one of their engineers. The platform we leave behind is only valuable if their team can run it.
  • Set the rolloff date at the start. Knowing when we leave forces real handoff, not consulting drift.
/ NEXT STEP

A platform problem that's been stuck for a quarter?.

Tell us where you are. 30-minute call, no NDA needed for the first conversation. We'll come back with the team shape and the first thing we'd ship.

Book a call or hello@nearcore.it