Roger Gelonezi

00 / Career highlights

Engineering impact, in production

Six results that show how I work: backend first, measured in production, delivered with a team.

Distributed platform

Architected and evolved a cloud-native ecosystem of 25+ C#/.NET APIs and services on AWS EKS, processing 200,000+ requests per day.
Legacy modernization

Led the progressive replacement of a complex legacy judiciary integration platform over about four years, operation by operation, with zero-incident client cutovers.
External integrations

Integrated ~20 heterogeneous Brazilian judiciary systems through REST, SOAP, asynchronous messaging and shared protocol contracts.
Cloud efficiency

Cut AWS resource reservations by ~40% and eliminated API throttling by tuning Kubernetes autoscaling from production metrics.
Technical leadership

Mentored five engineers from intern to senior and set up the architecture documentation, onboarding and engineering practices around them.
Applied AI

Raised a production legal-document classifier from ~40% to 80%+ accuracy while expanding it from ~12 to 140+ labels.

01 / Intro

Meet me in three minutes

Unscripted, in English (C1): where I come from, the system I'm proudest of, and how I lead.

Chapters

“I've always been a tech lead, even at the beginning of my career. It's something I carry with a lot of pride.”

Read the transcript (lightly edited for readability)

Hi, my name is Roger. I'm a senior software engineer, and I started my career in 2001. So yes, I have a lot of experience behind me: I've worked with many kinds of companies and industries, and built software for many kinds of architectures and server environments. I'll focus on my latest projects, or this would be a half-hour video.

I can talk a little about a set of microservices: 25 microservices in a fully scalable architecture, built to replace legacy systems. Each microservice replaced, one by one, an operation of the legacy monolith. It was a fully agile environment, so we could validate step by step, in real time, and every step taught us something about reading and writing documentation.

I was the main person responsible for understanding the legacy system and designing the documentation and the architecture of the new system: how one would replace the other. I was the point of contact working remotely inside the client. I followed up on every situation personally, and we had a really good set of metrics and observability, so we could prove the replacement was a full success, 100% okay, and release the new software.

I was also the main person responsible for onboarding and mentoring, from interns and junior developers to senior developers: setting up how the team works and how to collaborate with it. I've always been a tech lead, even at the beginning of my career. It's something I carry with a lot of pride, and I've received many compliments and thanks for it.

Hope we can talk again. Thank you so much.

02 / Featured system

Replacing a legacy monolith, one operation at a time

The integration platform I architected at Softplan. Internal products needed to talk to ~20 court and prosecutor systems that each spoke a different dialect and changed without notice. I gave them one API contract, built as ~25 .NET microservices on AWS EKS, that replaced the legacy platform over four years without a single incident at cutover.

~25.NET microservices on AWS EKS
200k+requests per day
~20judicial & prosecutor systems integrated
0incidents across client cutovers

Component

The whole platform

Consumers integrate once. Behind that single contract, stateless connectors absorb the constant change in external court systems, and each one ships, scales and fails independently.

Vertical Slice
Hexagonal
MassTransit
Saga

Design decisions

N-01
One contract, many systems
Consumers integrate once, against a five-operation API. Churn in external systems stays behind it.
N-02
Stateless connectors
One adapter per external system, deployed and scaled independently, so one slow court never drags down the rest.
N-03
Shared contracts in an SDK
Protocol versions (a national SOAP standard, plus REST) and per-system corner cases live in one versioned package.
N-04
Configuration over code
A catalog describes each external system, so onboarding a new court starts as configuration.
N-05
Sync and async paths
Request/response through the ingress; long-running work through RabbitMQ, with multi-step workflows as MassTransit sagas.
N-06
Replace, observe, retire
Vertical slices replaced legacy operations one by one. Elastic APM traces were the evidence behind each client cutover.

Generalized reference: product names and client specifics removed. Full board on Excalidraw ↗ Source ↗

03 / Scope

What I can own

Hand me one of these and I'll take it from analysis to production. Each links to the evidence.

Backend architecture

APIs, service boundaries, data flows and the contracts between distributed services.
See the platform →
Legacy modernization

Understanding existing behavior, planning incremental replacement and taking risk out of the migration.
How I modernize →
Integration platforms

Connecting heterogeneous third-party systems through stable contracts and resilient adapters.
Integration work →
Production reliability

Observability, CI/CD, troubleshooting, autoscaling and releases you don't have to hold your breath for.
Production evidence →
Technical leadership

Architecture decisions, mentoring, RFCs and ADRs, onboarding and cross-team alignment.
How I lead →
Applied AI

Agents and ML capabilities in real products, without weakening engineering discipline.
Applied AI →

04 / Experience

20+ years, from VB6 to agentic systems

Most recent first. The numbers are from production, not slides.

Softplan

May 2021 — Present · Legal-tech, Brazil
1. Senior Software Engineer — Backend .NET & Architecture
  
  May 2026 — PresentRemote
  - Took end-to-end ownership of the C#/.NET backend of an AI-first product; restructured the roadmap with Product into small, parallel workstreams shipped through continuous delivery, code review and automated tests.
  - Set engineering standards and architectural direction through RFCs and ADRs, while staying hands-on in development and architecture reviews.
  - Building a legal assistant on Microsoft Agent Framework + AG-UI that answers questions, summarizes case files and helps draft legal documents, with human oversight at key decisions.
  - C# / .NET
  - APIs
  - RFC · ADR
  - Code review
  - MAF
  - AG-UI
2. Senior Software Engineer — Machine Learning Product
  
  Jan 2026 — May 2026Brazil
  - Took over a production ML product for the Brazilian judiciary and raised classification accuracy from ~40% to 80%+, scaling the classifier from 12 to 140+ labels and automating case distribution at a public prosecutor's office.
  - Redesigned the C# serving API toward a single multi-tenant, high-confidence model; made training reproducible with MLflow and Qdrant.
  - Documented the ML pipeline end to end and mentored engineers to take over operations and client integrations.
  - C# serving API
  - MLflow
  - Qdrant
  - Model lifecycle
  - Python
3. Microservices Integration Engineer
  
  May 2021 — Jan 2026Lençóis Paulista, SP
  - Architected a cloud-native platform of 25+ C#/.NET APIs on AWS EKS (Vertical Slice + Hexagonal, shared SDK, MassTransit + Saga) handling 200,000+ requests/day at 99% uptime, with full CI/CD, automated testing and production observability.
  - Integrated ~20 Brazilian judiciary systems over REST and SOAP with real-time sync; replaced days of manual validation with Robot Framework pipeline gates.
  - Cut AWS resource reservations by 40% and eliminated API throttling with an AI-driven routine that tunes Kubernetes autoscaling from Grafana metrics.
  - Led the four-year replacement of the legacy platform with zero-incident cutovers; mentored five engineers from intern to senior in a GovTech domain.
  - .NET
  - MassTransit
  - RabbitMQ
  - Saga
  - AWS EKS
  - Kubernetes
  - Helm
  - Elastic APM
  - Grafana
  - Robot Framework
Eris Sistemas

Oct 2020 — May 2021 · Bauru, SP
1. Full-stack Engineer
  - Analyzed and planned the modernization of a VB6 + .NET Framework platform (SQL Server) for book publishers and bookstores: mapped ~60 functional areas, their business rules and dependencies, and defined a phased path to C#/.NET Core services and APIs.
  - Implemented the first migrated domain (Orders) on .NET Core and guided an intern through validation; a dedicated team later continued the approach.
  - Designed and built a React Native self-checkout with PagSeguro payments for a major São Paulo bookstore. PagSeguro later cited it as a benchmark: their first React Native self-checkout.
  - C# / .NET Core
  - VB6
  - SQL Server
  - PagSeguro
  - React Native
Boardgames

May 2016 — Sep 2020 · São Paulo
1. E-commerce Technology & Operations Specialist
  - Owned e-commerce technology end to end: OpenCart launch, Magento migration, payment and shipping integrations (Mercado Pago, Correios) and marketplace integrations (Mercado Livre, Amazon, B2W, Magalu).
  - Built the internal reporting app behind purchasing, promotions and pricing decisions; as revenue grew to ~R$1.5M/year, also took on operational leadership, including hiring and managing staff.
  - Integrations
  - Payments
  - Marketplaces
  - Reporting
Lecom

Mar 2014 — Apr 2016 · Bauru, SP
1. Full-stack .NET Engineer & Technical Lead
  - Tech lead for the City of Bauru's GovTech tax platform, used by 60,000+ citizens to review, update and pay municipal taxes online.
  - Ran the knowledge transfer that let the city's IT team own the codebase: bi-monthly training on architecture, design patterns and code standards, plus a city engineer embedded in our team.
  - C#
  - ASP.NET MVC
  - SQL Server
Earlier experience

2002 — 2014
- MODEC · Full-stack Engineer · 2013–2014. A C#, ASP.NET and SQL Server application replacing spreadsheet-heavy workflows for oil-extraction measurement reports to the Brazilian government (ISO 10012).
- Solução Informática · Full-stack Engineer · 2007–2013. Modernized a Visual Basic product to C#/.NET; development, maintenance, customer support and client-facing pre-sales.
- Correios · Programming Intern · 2002–2003. VB6, SQL Server and Crystal Reports on internal business systems.

05 / Expertise in depth

Modernize, integrate, operate

Most of my work happens after the greenfield phase: inside systems that are already running a business.

E-01 · Legacy modernization

Modernizing systems without stopping the business

Put a contract in front of the old system, replace it piece by piece, and prove each step in production before retiring anything. Analysis and documentation come first; big-bang rewrites don't.

Softplan: a judiciary integration platform replaced operation by operation over ~4 years, client by client, with zero-incident cutovers and final legacy retirement.
Eris Sistemas: analyzed ~60 functional areas of a VB6 + .NET Framework platform and planned its move to C#/.NET Core, delivering the first domain.
Earlier: Visual Basic to C#/.NET at Solução Informática; spreadsheet workflows to a web application at MODEC.

Understand legacy behavior and business rules
Map dependencies and migration risks
Define the replacement contract
Migrate incrementally, one operation at a time
Run old and new side by side during the transition
Validate behavior and monitor production
Cut over client by client, with rollback in mind
Document, then retire the legacy component

E-02 · Distributed systems

Distributed systems & integrations

Independently deployed services that have to coordinate reliably across external dependencies they don't control. That's where I've spent the last five years, and much of the decade before it.

C#
.NET
ASP.NET Core
REST
SOAP
Microservices
RabbitMQ
MassTransit
SQL Server
PostgreSQL
AWS
Kubernetes
CI/CD

Backend APIs in C#/.NET and ASP.NET Core, from layered MVC to Vertical Slice + Hexagonal
REST and SOAP, including a national interoperability standard in several versions
Third-party APIs: payments (PagSeguro, Mercado Pago), shipping, marketplaces
Asynchronous messaging with RabbitMQ and MassTransit; Saga workflows for multi-step processes
Configuration-driven connectors over a shared SDK of protocol contracts
Resilience: slow or failing external systems isolated behind async paths and independent scaling
Data access on SQL Server and PostgreSQL with EF Core
Distributed tracing and production troubleshooting across services

E-03 · Production

Production is part of the design

I don't hand systems over the wall. Observability, tests and scaling behavior are designed with the feature, and they're how I know a change is safe to ship.

0incidents across client cutovers
−40%AWS reservations, throttling eliminated
Days → minutesmanual validation replaced by pipeline gates

Logs, metrics and traces with Elastic APM, Elasticsearch and Kibana; dashboards and alerts in Grafana; OpenTelemetry
Automated tests as gates: xUnit, integration tests with Testcontainers, Robot Framework in the pipeline
CI/CD on GitLab and GitHub Actions, deploying to Kubernetes with Helm
Autoscaling tuned from data: an AI-driven analysis of Grafana metrics drove the Kubernetes settings
Cutover monitoring: metrics and traces as the go/no-go evidence for each client
Production troubleshooting across services and external dependencies

06 / How I work

Five principles I ship by

P-01
Agents draft, I decide

AI tools inside guardrails: specs first, small diffs, tests, and every change reviewed before merge. The architecture and the accountability stay mine.
P-02
Decisions in writing

RFCs and ADRs turn hallway decisions into durable context. New engineers learn why, not just what, and architecture stops depending on who was in the room.
P-03
Observable, or it isn't done

Metrics, traces and alerts ship with the feature. That's what turned zero-incident cutovers into a measurement instead of a hope.
P-04
Build teams that outgrow me

Mentoring from intern to senior, onboarding that sticks, and knowledge transfer deep enough that a client's own team can take over the codebase.
P-05
Own the outcome, not the ticket

Architecture, implementation, deployment and production behavior are one engineering problem. I'd rather own it end to end than hand work off at the edge of a ticket.

Technical leadership

Leverage, not heroics

Staff-level work is mostly decisions other people can build on, and engineers who don't need me in the room. I stay hands-on while doing it.

Architecture ownership: Technical reference for the judiciary platform's architecture and its replacement; RFCs and ADRs for current work at Softplan.
Mentoring: Five engineers from intern to senior; engineers prepared to own a production ML model and its client integrations.
Onboarding & documentation: Led onboarding into a complex distributed-systems domain; documented the architecture, system behavior and the ML pipeline.
Code review & delivery: Code reviews, automated tests and release discipline behind continuous delivery.
Working with Product: Decomposed roadmap initiatives into smaller, parallel workstreams the team could ship independently.
Knowledge transfer: Training and an embedded engineer so the City of Bauru's IT team could own its tax platform.
Cutover coordination: Personally owned critical client cutovers through final legacy retirement.

07 / Projects

Things I've built

Open-source work you can read, plus professional products I can describe but not show.

Professional · Softplan · 2021–2026 · GovTech

Judiciary integration platform

The cloud-native platform behind the featured system: it replaced a legacy system over four years and keeps ~20 Brazilian judiciary systems in real-time sync.

200,000+ requests/day · 99% uptime, with zero-incident client cutovers.

Architecture: 25+ C#/.NET APIs · Vertical Slice · Hexagonal · shared SDK
Messaging: MassTransit · Saga · event-driven · real-time notifications
Platform: Kubernetes on AWS · S3 document storage · full CI/CD
Quality: Robot Framework pipeline gates · production observability

See the architecture ↑ Excalidraw board ↗ Proprietary: no public code

Now building · Open source

MyMechanicShop

A realistic domain for practicing backend architecture: a modular monolith where every business area is its own ABP module with its own database.

Architecture: Modular monolith · DDD · module-per-database
Backend: .NET 10 · ABP · EF Core · PostgreSQL · OpenIddict multi-tenancy
Quality: CodeQL · SonarCloud quality gate · Angular client

Live (early stage) ↗ Code ↗

FIAP postgrad · Team of four · Open source

Mecânica Hermes

A distributed backend for a mechanic shop, built by a four-person team across the postgrad's Tech Challenge phases: vehicle intake, quotes approved by the customer via signed webhook, Mercado Pago payment and vehicle release.

AI-assisted delivery. A phase scoped for a two-month term implemented in ~20 hours, with the architecture and every change reviewed by me.

Services: 3 .NET 10 microservices + Lambda · bounded contexts · shared SDK of 6 NuGet packages
Messaging: RabbitMQ · MassTransit · Saga · transactional Outbox · idempotent consumers
Data: PostgreSQL · MongoDB
Cloud: Terraform · EKS · RDS · Cognito · API Gateway · Kustomize
Quality: ≥ 80% coverage gate · Robot Framework E2E · OpenTelemetry → New Relic · ADRs

Project wiki ↗ Repositories ↗

Professional · Softplan · 2026

Legal AI assistant

A C#/.NET assistant over lawsuit documents: agent workflows answer questions, summarize case files and help draft legal documents, with a human approving key decisions.

Stack: C# / .NET · Microsoft Agent Framework · AG-UI
Focus: Agent workflows · document context · human oversight

More in Applied AI ↓Proprietary: no public code

Professional · Softplan · 2026

Legal document classifier

A production ML classifier that routes cases at a public prosecutor's office, replacing manual triage.

~40% → 80%+ accuracy, scaled from 12 to 140+ labels.

Stack: C# serving API · MLflow · Qdrant · Python
Focus: Multi-tenant model · reproducible training

More in Applied AI ↓Proprietary: no public code

Professional · Eris Sistemas · 2021

Bookstore self-checkout

A self-checkout with PagSeguro payments and a .NET backend for a major São Paulo bookstore. PagSeguro later cited it as a benchmark: their first self-checkout built in React Native.

Stack: .NET backend · PagSeguro · React Native

Proprietary: no public code

08 / Applied AI

Applied AI, not AI theater

AI strengthens how I build backend systems; it doesn't replace engineering judgment. Three areas, all in production or daily practice.

A-01

AI-enabled product engineering

A legal assistant built in C#/.NET on Microsoft Agent Framework and AG-UI. Agent workflows over case-file context answer questions, summarize case files and help draft legal documents.

Humans approve key decisions; client details stay confidential.

C# / .NET
Microsoft Agent Framework
AG-UI
Human oversight

A-02

ML product ownership

Owned a production legal-document classifier: accuracy ~40% → 80%+, labels ~12 → 140+. Redesigned the C# serving API toward a multi-tenant model and made the training lifecycle reproducible.

Python here is a tool for the ML pipeline, not my primary language.

C# serving API
MLflow
Qdrant
Python

A-03

AI-assisted software engineering

Claude Code, GitHub Copilot, Cursor, Codex and MCP as accelerators, inside a discipline:

Specs first, work decomposed into small tasks
Architecture decisions made by me
Small diffs, backed by automated tests
Every generated change reviewed and validated

Result on a FIAP phase: ~20 hours of AI-assisted implementation, every change human-reviewed.

09 / Stack

Bill of materials

Grouped by how deep it goes. Highlighted = daily drivers.

Backend primary: C#
.NET
ASP.NET Core
EF Core
REST APIs
MassTransit
Architecture primary: Distributed systems
Microservices
DDD
Event-driven
Vertical Slice
Hexagonal
Clean Architecture
Modular monolith
CQRS
Saga
RFC
ADR
Cloud & platform: AWS
EKS
Kubernetes
Docker
Helm
Terraform
GitLab CI
GitHub Actions
Messaging & integration: RabbitMQ
MassTransit
Kafka
AWS SQS
Temporal
REST
SOAP
Data: SQL Server
PostgreSQL
MySQL
MongoDB
DocumentDB
Redis
Observability & quality: Elastic APM
Elasticsearch
Kibana
Grafana
OpenTelemetry
New Relic
xUnit
Testcontainers
Robot Framework
SonarCloud
CodeQL
Security & identity: Keycloak
AWS Cognito
OAuth2 · OIDC · JWT
Vault
AI / ML secondary: Microsoft Agent Framework
AG-UI
RAG
Qdrant
MLflow
Semantic / vector search
Python
AI-assisted engineering: Claude Code
GitHub Copilot
Cursor
Codex
MCP
Spec-driven development
Agentic workflows
Front-end & mobile supporting: React Native
Angular
ASP.NET MVC

10 / Writing

Notes from the field

Short posts on Staff engineering, architecture, AI-assisted development and remote work.

Oct 2026 · LinkedIn

The job descriptions didn't get easier. I grew.

In 2020 a Staff Engineer job description looked like a hiring plan for a whole team. Today I read the same lists and think: done that, worked with that, and I can bring more.

Staff engineering
Career growth

May 2026 · LinkedIn

Learning the Saga pattern in a weekend

Implementing Saga with MassTransit in C# for about $30 in tokens. My rule: make the AI report what it did and why, like a teacher, so you learn the architecture and not just the prompts.

Saga
MassTransit
Engineering judgment

Sep 2026 · LinkedIn

The nightmare where I worked from an office

A bad dream about commutes and constant interruptions, and what it taught me about focus, my own space and time with family. Quality of life is the raise I won't trade back.

Remote engineering
Focus

May 2026 · LinkedIn

From Windsurf to Claude Code

Why I switched AI coding tools after a year, and the night I sketched a feature at the gym and came home to find it built.

AI-assisted engineering
Developer tooling

All posts on LinkedIn ↗

11 / Education

Still studying the craft

Postgraduate · Lato Sensu · 360 h

Software Architecture

FIAP · São Paulo · Aug 2025 — Jul 2026

91.8/100 average across 27 modules, all passed

Modules · highlighted = top mark (96)

Microservices I & II
Saga Pattern
Microservice Resilience
Serverless
API Gateway
Monitoring & Access I & II
Data Engineering
Graph Databases
High-Performance Teams
IT Communication & Negotiation
Data Privacy & LGPD
Software Architecture
DDD
Secure Development
Architecture Documentation
Software Quality
LeetCode
Clean Architecture
Kubernetes I & II
Terraform
DevOps (IaC & CI/CD)
GitHub Actions
Docker
OpenTelemetry

B.Sc. Information Technology

UNIVESP · São Paulo State Virtual University · 2022 — 2025
Bachelor's, Business Administration

ITE · Instituição Toledo de Ensino · 2006 — 2008
Technical Diploma, Software Programming

CTI “Prof. Isaac Portal Roldán” · 2001 — 2003

Languages

EnglishC1 · full professional
PortugueseNative

12 / Contact

Let's build something that scales.

I'm open to Senior and Staff Backend Engineer and Software Architect roles, remote from Brazil (UTC−3), as a full-time employee or B2B contractor. LinkedIn is the fastest way to reach me; my contact details are open to recruiters there.

Message me on LinkedIn View résumé

The whole platform

One contract, many systems

Stateless connectors

Shared contracts in an SDK

Configuration over code

Sync and async paths

Replace, observe, retire

Backend architecture

Legacy modernization

Integration platforms

Production reliability

Technical leadership

Applied AI

Softplan

Senior Software Engineer — Backend .NET & Architecture

Senior Software Engineer — Machine Learning Product

Microservices Integration Engineer

Eris Sistemas

Full-stack Engineer

Boardgames

E-commerce Technology & Operations Specialist

Lecom

Full-stack .NET Engineer & Technical Lead

Earlier experience

Modernizing systems without stopping the business

Distributed systems & integrations

Production is part of the design

Agents draft, I decide

Decisions in writing

Observable, or it isn't done

Build teams that outgrow me

Own the outcome, not the ticket

Leverage, not heroics

Judiciary integration platform

MyMechanicShop

Mecânica Hermes

Legal AI assistant

Legal document classifier

Bookstore self-checkout

AI-enabled product engineering

ML product ownership

AI-assisted software engineering

The job descriptions didn't get easier. I grew.

Learning the Saga pattern in a weekend

The nightmare where I worked from an office

From Windsurf to Claude Code

Software Architecture

Let's build something that scales.