AI Engineering in Production
We build AI agent systems that survive production.
Most AI projects stall between a working demo and production. We build the agent systems that make it, and the cloud, data, and observability foundation that keeps them running at scale.
What we build
Production AI is the work. Everything else is the foundation that keeps it running once it's live.
AI Engineering
Production AI systems, not demos: Domain Agent Harnesses, RAG and Knowledge AI Databases, Horizontal AI Agent Scaling, and the AI Ops that keep them running. From first prototype to a system your own engineers can own.
The foundation
Cloud & Infrastructure
Architecture, migration, and cost optimization across AWS, GCP, and Azure. Infrastructure as code and CI/CD pipelines built to scale with the team that inherits them.
Observability
Instrumentation, alerting, and dashboards that catch problems before customers do. Logs, metrics, and traces wired into a system your team can actually read at 3am.
IoT
We connect the physical world to your infrastructure with dedicated IoT hardware and specialized protocols. Real-time data capture and processing through to the user-facing interaction layer, all integrated into a single system.
Client work
A culinary platform that needed AI in two places at once: inside its own operations, and in the product its customers use. We built the infrastructure for both and got it running at scale.
- Agent infrastructure
- RAG
- Customer-facing AI
- Cloud
An engineering team that needed senior AI capability without slowing down. We embedded directly and shipped AI Engineering systems into production alongside them.
- AI Engineering
- Embedded delivery
- Production systems
- yummyFOOD TECHNOLOGY
A point-of-sale platform for food retail, built from the ground up. We combined IoT at the edge with AI Engineering to modernize how the sale actually happens.
- End-to-end platform
- IoT / edge
- AI at the point of sale
How I work
I'm Felipe Romani. I build production systems: Go and Python services on Kubernetes, event streaming, and the orchestration that keeps them reliable. For the last few years that work has been AI, getting language models to do real jobs inside real products, at scale.
Most AI projects don't fail at the model. They fail at everything around it: orchestration, retrieval, evaluation, cost, and what happens when a step breaks at 2am. That's not a prompting problem, it's a systems problem, the same discipline that makes any distributed system survive production, applied to a new kind of workload. I build the whole stack, and I build it so your team can own it after I'm gone.
Felipe Romani on LinkedInContact
No sales team, no ticket queue. Email goes straight to the engineers who would work on your project.
We reply from Brazil, business hours ART/BRT. English and Portuguese both work.