I've been building with AI since 2020 — as one of OpenAI's earliest closed beta testers, long before the technology entered the mainstream.
Every engagement starts the same way: by deeply understanding how a business actually operates. That immersion — combined with years of hands-on experience across the full AI stack — is how I deliver systems that don't just work technically, but fit the reality of the people who use them.
Rodrigo's depth of knowledge across AI systems is unlike anything I've encountered — and I've spoken with teams at Oracle, AWS, and every major vendor. He doesn't just understand one corner of this space; he sees the full picture, from infrastructure to models to business application. That's extraordinarily rare.
What set Rodrigo apart was how quickly he understood our business. Within weeks he was speaking our language — regulatory nuances, market dynamics, all of it. He didn't just bring AI expertise; he became an expert in what we actually do.
On working with Rodrigo:
We'd talked to consultants who understood technology and consultants who understood our ecosystem, but never both. He went deep into years of our history and built something that made all of it useful.
A closer look at how AI strategy becomes production reality — from architecture through deployment.
Casa Castilho Engenharia
Casa Castilho Engenharia processes hundreds of CLCB and AVCB fire safety certificates for the São Paulo Fire Department each year — a workflow that demands deep regulatory interpretation, parsing of architectural drawings across multiple technical formats, and meticulous assembly of compliance dossiers for government submission. Every certificate carries legal weight: errors or omissions can mean project delays, fines, or safety risks. As the firm's client base grew, the fully manual process hit its ceiling — slow turnaround, inconsistent quality across team members, and no scalable path forward without multiplying headcount.
I started by spending weeks embedded in Casa Castilho's workflows — sitting with the team, understanding every decision point in the certification process, learning the regulatory nuances that no technical spec would capture. That immersion shaped every architectural choice that followed.
The system's intelligence layer combines multiple AI capabilities: optical character recognition for scanned documents, specialized parsers for DWG/DXF architectural files and IFC building information models, and a retrieval-augmented generation (RAG) pipeline that lets the team query São Paulo's fire code regulations in natural language — turning hundreds of pages of Instruções Técnicas into an interactive knowledge base.
What makes the system reliable for regulatory work is the deterministic rules engine underneath. While AI handles extraction and interpretation, compliance checks run through hard-coded rule sets that encode the exact requirements of each applicable regulation. Every decision is traceable — which rule was applied, which version, against which input — creating a full audit trail.
A custom orchestration layer coordinates the full AI pipeline: from document parsing through validation, gap analysis, dossier generation, and automated submission to the government's Via Fácil Bombeiros portal via RPA. The system even monitors the portal for UI changes that could break the automation — a pragmatic detail that reflects how production AI systems need to be resilient, not just functional.
Confidential — Insurance Lead Marketplace
A rapidly scaling insurance lead generation company was managing a high-volume operation spanning multiple media buying channels, aggregated lead feeds, owned campaigns, and programmatic acquisition — with a distributed team across three countries. Their lead lifecycle relied on a constellation of platforms for distribution, call routing, analytics, validation, fraud detection, and intent scoring.
The business had a clear opportunity to use AI for smarter lead scoring, quality prediction, and routing optimization — but operated in a regulatory environment where consumer data handling isn't just a compliance checkbox, it's an existential risk. They needed an AI strategy that could deliver real competitive advantage without any consumer data ever leaving their infrastructure.
Before touching any technology decisions, I spent time understanding the business from the inside out — the economics of a lead, the margin structure at each handoff, the regulatory exposure at every data touchpoint. That context turned out to be as important as the technical design.
I mapped the end-to-end lead lifecycle and audited over twelve integration points to understand data flows, quality signals, and decision logic at each stage.
The defining architectural decision was building entirely on-premise. I designed and specified custom hardware for local large language model inference — purpose-built machines optimized for the specific model architectures and workloads their business requires. This isn't a generic "run AI locally" approach; it's infrastructure engineering tuned to production throughput, latency requirements, and cost efficiency for their specific use case.
The AI roadmap covers three core capability areas: predictive lead scoring models trained on their proprietary conversion data, automated quality assurance pipelines that catch issues before leads reach buyers, and intelligent call routing that optimizes match quality in real time. All designed to run within their own walls — no cloud AI APIs, no data leaving the building.
Endeavor Brazil
Endeavor Brazil sits at the center of Latin America's most influential entrepreneurship ecosystem — running over twenty acceleration programs annually, cultivating relationships with hundreds of high-growth founders and C-level mentors, and connecting them through a network that has shaped some of the region's most significant companies. Over twelve years, the organization had accumulated an extraordinary body of institutional knowledge: detailed meeting notes with founders, mentor feedback, program evaluations, strategic assessments, and ecosystem intelligence across every major sector.
But this knowledge was locked — scattered across databases, documents, and the memories of long-tenured team members. When a program director needed insight about a founder's trajectory, or a mentor wanted context on a portfolio company's challenges, the answer required manual digging or knowing the right person to ask. Endeavor was sitting on one of the richest entrepreneurial datasets in Latin America and had no way to tap into it at scale.
I immersed myself in Endeavor's world — understanding not just the data structures but the relationships, the program dynamics, the way institutional knowledge actually flows (and gets lost) in a network organization. That depth of understanding shaped a system designed for how people actually work, not how a spec document imagines they do.
This was one of the earliest practical implementations of retrieval-augmented generation (RAG) in the Brazilian market — built at a time when the technology was still emerging and the playbooks hadn't been written yet. Having been a closed beta tester for OpenAI since 2020, I brought both deep familiarity with the underlying model capabilities and the practical judgment to know where AI would deliver real value versus where it would disappoint.
Twelve years of meeting notes, founder interactions, mentor feedback, and program records were processed into a unified knowledge layer — indexed, semantically structured, and made accessible through natural language queries. The result is an AI assistant that allows anyone in the organization to tap into over a decade of accumulated insight as naturally as asking a colleague. A program director preparing for a founder meeting can surface every prior interaction, recommendation, and strategic note in seconds. A mentor can understand a company's full history with Endeavor before walking into a room. Leadership can identify patterns across cohorts, sectors, and programs that would have taken weeks to research manually.
The system transforms institutional knowledge from a static archive into a living, queryable intelligence layer — and proves that AI's most powerful applications aren't always about generating new content, but about making existing knowledge accessible and actionable.
I work with companies navigating complex AI challenges — from early strategy through production deployment. If you're looking for someone who builds, not just advises, let's talk.