From tightly coupled architecture to a platform built for scale
Leadership of a transformation combining performance, reliability, architecture, and operating model for products including Limpa Nome, eCred, and Premium.
The challenge
Strategic products grew rapidly, but critical integrations, mainframe dependencies, and duplicated capabilities created a tightly coupled architecture. Marketing campaigns frequently caused incidents, and some journeys took almost four minutes.
Before
- Reactive and expensive scaling during incidents
- Unclear ownership between platform and vertical teams
- Heavy operations in the synchronous customer journey
- Recurring campaign failures
Target state
- Centralized and reusable shared capabilities
- Clear end-to-end ownership
- Observable asynchronous processing
- Capacity planned before campaigns
What I personally led
Shared platform
Direct leadership of eight engineers responsible for the foundation used by Limpa Nome, eCred, and Premium.
Cross-functional coordination
Alignment across SRE, DevOps, infrastructure, product, security, and business-unit leaders.
Direction and priorities
Platform strategy, critical journeys, and requirements for LGPD, fraud prevention, access, performance, and reliability.
The key decision
Map
Latency, mainframe, integrations, duplication, and ownership.
Prioritize
Impact, risk, scale, and reuse potential.
Decouple
Caching, Go, Redis, queues, retries, and observability.
Operate
Capacity planning and contingency before campaigns.
Execution across two main workstreams
1 · Positive Credit Registry
Scalable caching, Redis-based enrichment, and cloud services in Go reduced dependency on remote processing in São Paulo and the mainframe.
Delivery: approximately 1 month
2 · Limpa Nome
Heavy operations were removed from the synchronous path and replaced with asynchronous processing, safe retries, better failure handling, and improved observability.
Transformation: approximately 2 months
3 · Campaigns
Product, SRE, DevOps, and infrastructure began reviewing traffic, scale, monitoring, and contingency plans in advance.
Model: continuous planning
Measurable outcomes
Scale proven
Peaks of approximately 20,000 accesses per second for ten minutes.
Stable campaigns
Production stopped failing during marketing campaigns.
Cost avoided
Eliminated reliance on emergency scaling of up to ten additional machines for ten hours.
Approximate values based on project records.