Felipe Xavier
Fine-tuning, evaluation and serving of LLMs (Transformers); NLP pipelines; multi-agent orchestration; and the application architecture that takes all of it to production at scale.
Autonomous multi-agent system in production
I designed and shipped Becomex's autonomous multi-agent system to production — a supervisor/worker topology where orchestrator agents decompose work and route it to specialized workers running domain fine-tuned LLMs. A RAG pipeline over a Weaviate vector store grounds every step with Chain-of-Thought, while self-reflective ReAct loops detect and recover from their own errors. It runs on Azure Kubernetes with Redis-backed state and end-to-end observability — and cut operational effort by 50%.
Engineering at the world's largest consultancy
At the world's largest technology consultancy I owned integration architecture for global brands. For one of Latin America's largest telecoms (10M+ users) I designed fault-tolerant microservices for business-critical flows, optimized APIs under heavy load and hardened security across auto-scaling AWS with monitoring — and set the architecture patterns the teams built on.
From software engineering to enterprise
Wiipo · Software Engineer · Mar — Sep 2021
Took a fintech platform from MVP to production: led post-acquisition integration, shipped the React Native app and the React.js B2B platform, and built a serverless backend on AWS Lambda + DynamoDB behind an automated credit-decisioning engine.
Neogrid · Software Developer · Mar 2019 — Mar 2021
Enterprise systems for global supply chain on Java/C#: fiscal-compliance engines and internal automation, workloads on Azure, and contributions to new-product architecture.
Who I am
I'm an AI/ML engineer with 8+ years across software engineering, solution architecture and data science — the blend that carries an idea from a research notebook to a system that survives production. I architect autonomous multi-agent systems — supervisor/worker topologies over fine-tuned LLMs, retrieval-augmented generation grounded by Chain-of-Thought, and self-reflective ReAct loops that catch and correct their own failures — on the engineering discipline I built shipping distributed systems for 10M+ users: event-driven microservices, Kubernetes and rigorous failure-mode thinking.
- Focus
- LLMs · NLP · Transformers · Agents
- Foundation
- 8+ yrs · distributed systems
- Cloud
- AWS · Azure (AKS) · Kubernetes
- Based in
- Joinville, Brazil · remote
How I build
Machine Learning · NLP · LLMs
Cloud & Infra
Engineering
Architecture & Leadership
Let's talk
Open to opportunities in AI/ML, autonomous systems and software architecture.
© · Felipe Xavier · Staff AI/ML Engineer · Santa Catarina, BrazilThanks for making it this far.
— Felipe Xavier · Staff AI/ML Engineer