João Gabriel de Barros Rocha

- Sole engineer on a production real-estate marketplace that aggregates listings from roughly 1,000 broker sites into a PostgreSQL dataset of over 100,000 properties, owning architecture, delivery, deployment and production support for a remote, English-speaking team working in Scrum. - Built the broker subscription and payments flow end to end with Stripe, covering checkout, plan selection, billing management and reliable webhook handling, along with the lead-routing and email notifications that subscribed agencies pay for. - Led the search performance work that took homepage load from 8 seconds to 1.3 and roughly halved response times across the platform, through indexing, caching, materialized views and moving heavy ranking logic from the application into the database. - Owned PostgreSQL data modeling and integrity: schema design, indexing and query tuning on live tables, zero-downtime migrations, and resolving production incidents around slow queries and duplicated or inconsistent records. - Redesigned the crawler into a staged extraction, enrichment and validation pipeline, improving data quality with image-fingerprint duplication detection across agencies, geocoding and coordinate validation, Redis-backed job queueing, and an LLM step that filters out stock and marketing photos. - Migrated production infrastructure from Azure to Hetzner Cloud, cutting hosting costs by roughly 75%, and set up the engineering baseline: GitHub with branch protection and pull-request reviews, automated deploy pipelines, daily database backups and a Pytest suite of 50+ test modules.

- Developed the initial version of the ES+ Criativo and HUBES+ systems for the Espírito Santo Secretariat of Culture using a Java stack based on Jakarta EE, JPA and JSF. - Migrated the platform to a PHP and WordPress stack, engineering custom plugins to extend core functionality and maintaining both systems in production.

HereToSlay

January 2026

Built an event-driven backend where gateway, lobby and game-engine services communicate only through an Apache Kafka cluster (3 brokers, KRaft mode), with no direct HTTP or RPC calls, allowing each service to scale independently as a replicated consumer group.