// principal / staff software engineer

Pablo Felipe

Financial Systems Architect · Distributed Systems · AI-Integrated Platforms

20+ years designing mission-critical financial platforms at global scale. Currently architecting fiscal middleware running in 25 countries with 10+ active tax regimes in production, where an error in the tax calculation engine is a compliance failure, not a bug report.

Portrait of Pablo Felipe
0 Years of Experience
0 Countries Served
0 Performance Gain
0 Fewer Incidents

Who I am

I'm a Principal/Staff-level backend engineer and architect focused on enterprise financial systems. My specialty is designing platforms that need to be correct before they're fast: transactions, regulatory compliance, and data integrity in production.

I've led the architectural evolution of a fiscal middleware platform at Oracle, serving operations in 25 countries across LATAM, EMEA, and Asia, with 10+ active tax regimes in production. The result: a 50% jump in transaction performance and a 40% drop in production incidents.

I'm fullstack-capable, but my passion is the backend: where data can't be wrong, where failure has real consequences, and where architecture genuinely matters.

Areas of focus

High-integrity financial systems architecture
Modular monoliths and distributed systems
Transaction processing and regulatory compliance
Cross-country fiscal platform integration
Eval-first RAG pipelines for regulated domains
Enterprise architecture with .NET, Java/Spring, and Go

Technical stack

⚙️

Backend & Architecture

C# .NET Python Java Spring Boot Node.js / TypeScript C++
🔗

Integration & APIs

REST GraphQL OpenAPI / Swagger WebSockets PWA / Push Notifications React 19 / TypeScript (Frontend) Microservices
🤖

AI / ML

RAG Pipelines Eval-First Harnesses ChromaDB Gemini API Multimodal (Gemini Vision) Provider-Agnostic LLM Layer
🗄️

Data & Messaging

PostgreSQL SQL Server Oracle MySQL MongoDB RabbitMQ
🐳

Infrastructure & DevOps

Linux Docker Kubernetes (kind) Terraform GitHub Actions Jenkins AWS (Lambda, RDS, ECS/Fargate) Azure (Container Apps) Firebase / Serverless
📊

Observability & Tooling

Prometheus / Grafana Agentic Workflows (Claude Code)

Flagship repositories

eval-first RAG

ncm-classifier-ai

A RAG pipeline that classifies Brazilian products into 8-digit NCM fiscal codes, grounded on the official TIPI table. Built eval-first: every architectural change is gated by a labeled suite tracking accuracy, calibration, latency, and cost, including the changes that didn't work. A deterministic verification gate routes low-confidence output to escalation. An OWASP Top 10 for LLM Applications audit found and fixed two real unauthenticated crash bugs and shipped a permanent regression suite; GenAI-semantic-convention OTel tracing never captures prompt or completion content.

RAG Hexagonal Architecture FastAPI ChromaDB Gemini Hybrid Search OWASP LLM Top 10 Audit
Python
event-driven microservices

easydora

A polyglot, event-driven e-commerce system built in Go, Spring Boot, and FastAPI, each used where it fits the workload. Every cross-service interaction flows through RabbitMQ topic exchanges; the Outbox Pattern guarantees an event is never silently lost; event contracts are validated against versioned JSON Schemas so producer/consumer drift is caught automatically.

Event-Driven RabbitMQ Outbox Pattern Contract Testing Kubernetes (kind) Observability (Prometheus/Grafana) OpenTelemetry / Jaeger Circuit Breakers
Java · Go · Python
container-first · multi-cloud

SmartCondo

A full-stack condominium administration platform, with ASP.NET Core 8 (REST + GraphQL via HotChocolate) on the backend, React 19 + TypeScript PWA on the frontend, PostgreSQL behind EF Core. The same Docker image deploys unmodified to Azure Container Apps or AWS ECS/Fargate through two independent Terraform modules, torn down and rebuilt from terraform apply alone to prove it. Real-time notifications run over native WebSockets by default, with zero AWS runtime dependency.

ASP.NET Core GraphQL React 19 TypeScript Terraform WebSockets Multi-Cloud Tenant-Isolation Audit
C#

Fiscal regulation × AI

Fiscal regulation and AI is a narrow intersection with very few engineers who've operated in both.

LLMs fail at fiscal classification out of the box: they hallucinate plausible-looking codes, can't express calibrated confidence, and leave no audit trail. The interesting engineering problem is the layer between the raw model and a regulated production environment: retrieval grounding, verification, structured output, confidence scoring, and human-in-the-loop design.

That's the problem ncm-classifier-ai is working on: treating AI as an engineering discipline, not a demo.

Professional track record

Principal Application Software Engineer
Oracle
Jul 2021 – Present

Technical lead and architect for a fiscal middleware platform serving 25 countries across LATAM, EMEA, and Asia, with 10+ tax regimes in production. Led a full redesign of an 8-year legacy fiscal interface into a modular JavaScript architecture, driven by Brazil's tax reform — defined the architecture, standards, and implementation strategy end to end, fully hands-on, with production rollout in 3 months. The rebuilt fiscal layer runs natively on Windows, Linux, and Android for the first time. Built a Jenkins CI/CD pipeline from scratch, now used across the LATAM fiscal team, and mentors engineers across distributed, multi-timezone teams without direct authority.

↑ 50% transaction performance ↓ 40% production incidents 25 countries · 10+ tax regimes
Senior Software Engineer
Oracle
Sep 2016 – Jun 2021

Led modernization of Brazil's fiscal system, establishing it as the regional standard for LATAM operations, and designed backend integration services for enterprise financial systems in highly regulated environments, enabling the platform's multi-country expansion.

↓ 60% critical bugs LATAM regional standard
Senior Application Development Specialist
Orizon
Mar 2016 – Sep 2016

Optimized a payment authorization system (C++/MongoDB), reducing transaction latency and cutting integration effort for new card formats.

↓ 35% transaction latency ↓ 40% integration effort
Development Coordinator
BLK Sistemas Financeiros
Jun 2014 – Jan 2016

Led a 9-person engineering team delivering financial software solutions, improving planning visibility and structured quality gates.

90% on-time delivery 9-person team
Senior Development Consultant
BTG Pactual
May 2013 – Jun 2014

Re-engineered middle-office systems, integrating legacy components (VB6/C++) with VB.NET.

↑ 30% operational uptime
Lead Software Architect
Agência Estado
Feb 2010 – May 2013

Designed real-time market-data integration middleware (C++/Java), unifying the solution across Linux and Windows environments, and built high-performance order-routing components supporting peak-load transaction volume. (Senior Development Analyst, Feb 2007 – Feb 2010.)

↑ 40% data processing speed ↑ 45% peak transaction volume

Education & certifications

Education

MBA in DevOps Engineering & Integration Architecture, FIAP, 2021
MBA in Project Management, FIAP, 2015
Specialization in Software Engineering, Universidade São Judas Tadeu, 2006
BTech in Data Processing, Universidade São Judas Tadeu, 2004

Certifications

PMP® (Project Management Professional), PMI, 2018
PSM I (Professional Scrum Master), Scrum.org, 2014
ITIL Foundation, EXIN

Continuous Learning

Agentic workflow patterns
AI/LLM & RAG

Let's talk

This site is a walkthrough of how I approach engineering: eval-first AI, event-driven systems, fiscal platforms at scale. If you've read through the projects above and want to dig into a design decision, a trade-off, or the reasoning behind any of it, I'm happy to talk it through.

Reach out if you'd like to discuss the portfolio, compare notes on architecture, or exchange ideas on fiscal systems and AI.