Luiz Paulo I.

I build back-ends for AI products, from the LLM agent down to the SQL it runs. Right now that's Agenticly, a crypto and finance AI assistant in production, where I work on the back-end and the data layer. Some of that work: - Text-to-SQL over BigQuery's public Bitcoin, Ethereum and Solana datasets. I found that a "current balance" table was really a change log, which took one answer from 20 Ethereum holders to the real 10,056, and I price every query with a dry run before it ships. - LLM pipelines in FastAPI and LangGraph on OpenAI models: routing each question, grounding answers on live market data (CoinGecko, Alpha Vantage), and keeping a fallback when a provider changes its API without notice. - An evaluation harness that measures cost and routing accuracy per route before any model change. On one route it showed a 10x cost cut. - A B2B API with API keys, shared credit pools and usage metering, on Firestore, Firebase Functions (TypeScript) and Stripe webhooks. - A Docker test rig that builds and boots the real production image before deploy, catching packaging bugs that unit tests can't see. Before moving to the back-end I spent 3+ years in e-commerce front-end (JavaScript, React, Liquid and PHP on Shopify, VTEX IO and WordPress). There I also structured an IT department, led a developer team and got the technical and SEO teams working together, so I know what the people consuming an API need from it. I work in English every day: tickets, code reviews, runbooks and QA reports. Stack: Python, FastAPI, LangGraph, OpenAI, BigQuery, SQL, Firestore, Firebase Functions, TypeScript, Docker, Google Cloud Run, GitLab CI.