Urban mobility · Paris + Montréal
Mobility scale-up: −30% cloud bill, same SLOs.
AWS bill doubled in 18 months without matching traffic growth. GreenOps audit, refactor, Karpenter, ARM64. Measured outcome.
KPI
−30%
cloud bill
Duration
9 mois
Team
4 engineers
Hub(s)
Paris + Montréal
An AWS bill that doubles in 18 months without traffic following: that's rarely growth, it's architecture debt being paid in cash.
The context
Urban mobility scale-up, 180 people, Paris and Montreal hubs, 4 million active users. Microservices platform in production for four years, platform team of 12 engineers. The CFO slammed the table: annual cloud bill exceeded 2.1M USD with 22% traffic growth over the same period.
The problem
- AWS spend x2 in 18 months for +22% traffic
- No team budget, no FinOps in place
- Systematic over-provisioning of EKS nodes (14% average CPU)
- MTTR at 47 minutes, noisy alerting, no distributed tracing
- Engineers unable to attribute a cost to a service
The approach
Six weeks of GreenOps audit, then seven months of incremental remediation. No big bang, no re-platforming. We started by measuring, then cut obvious fat, then rethought what needed rethinking.
The four workstreams
- Full observability: Prometheus, Grafana, Tempo, cost attribution per namespace via Kubecost
- Cluster Autoscaler migration to Karpenter: tight packing, spot first, aggressive consolidation
- ARM64 Graviton on 60% of stateless workloads after benchmarks
- Smart scheduling: night batches on interruptible spot, taints/tolerations reset
The stack
- Go 1.22, Kubernetes 1.29 on EKS
- Karpenter 0.34, Graviton2/3 (c7g, m7g)
- Prometheus, Grafana, Tempo, OpenTelemetry SDK
- Kubecost for attribution, Terraform for IaC
The results
- Cloud bill: -30% at iso-SLO over 9 months (-630k USD/year)
- MTTR: 47 min to 11 min (divided by 4)
- Average cluster CPU: 14% to 51%
- Traffic absorbed: +35% with no added capacity
- Estimated carbon footprint: -38% (client Scope 3 report)
« Abbeal taught us to look at our bill as an engineering signal, not as fate. We recovered budget to reinvest in the product. »
What we learned
Karpenter is a game changer but demands rigor on pod disruption budgets. ARM64 works on 60% of workloads, not 100%: some third-party C++ binaries resisted us for two months. The real sustainable lever is FinOps embedded in the team: we trained two internal relays so the drop holds after we leave.
// Read next
Luxury jewellery & watchmaking · Genève + Paris + Tokyo
Cartier: from audit to in-house private LLM.
Compass (front + back architecture audits), Mapper (watchmaking + jewellery product generator), competitive data ETL on BigQuery, and now a private LLM fine-tuned on Cartier's own infra. A long-term tech partnership on the data and AI stack of a luxury house.
LLM privé
fine-tuned on Cartier infra
Tier-1 bank · Paris
BNP Paribas: Reference Book PO, from React/Redux to product AI agents.
Three Abbeal engineers at the core of the PO Marketplace. React/Redux/Node platform initially, now augmented with a product RAG, Claude agents for PM assistance, and an event-driven Kafka layer to scale.
RAG
PO product catalog
Digital banking / FinTech · Tokyo (Tamachi)
Money Forward: data backbone of a brand-new digital bank in Tokyo.
Money Forward, a Japanese FinTech leader listed in Tokyo, partnered with a top-tier Japanese banking group to launch a brand-new digital bank built from scratch. Abbeal partners on the Data Engineering side: designing and operating the Data Hub (Databricks + Delta Lake + dbt + AWS Tokyo) serving JFSA reporting, AML, risk management.
Data Hub
from-scratch digital bank Tokyo