Herycklys G.

I design and build backend systems and integrations that have to be correct: payments, data syncs, AI evaluation, and services where a quiet bug costs money or trust. Senior engineer, 10+ years, mostly Go and Node.js. Based in Brazil (UTC-3), overlapping most of the US business day. Where that experience comes from: • Deel (current): EMET, Deel's public AI lab for model evaluation and training. I work on the platform where domain experts define how model answers are scored; its benchmarks and data train Deel's own LLMs. Before that, the integrations layer connecting Deel's payroll and HR platform to the systems customers already run. • Nubank (NuPay): refund, settlement and reconciliation flows for a payment platform serving 200+ merchants, including iFood, Uber and Amazon. Balance correctness and idempotency across asynchronous services. • Imgur (3M+ daily users): rolled out a content-moderation classifier that caught ~40% more harmful content, proven by running old and new models side by side in production for 3 months; a sitemap system that processed 6B+ entries in 3 months; PHP-to-Go and EC2-to-Kubernetes migrations that cut infrastructure cost. • Here on Upwork: Go connectors for an Airbyte-based data platform across three contracts in a year, an internal CRM taken from proof of concept to production, and an automated monitoring and alerting system. What I can help with: • System design and architecture reviews: data models, service boundaries, failure modes, and a written plan your team can execute. • API integrations and connectors: third-party APIs, webhooks, syncs and ETL, with retries, idempotency and reconciliation so records are not silently lost or duplicated. • AI inside real systems: LLM features with evaluations before launch, idempotent tool calls, per-customer cost limits, and a fallback when the model is wrong. I also design MCP servers and APIs that give AI agents safe access to your systems. • Backend services in Go or Node.js, built with tests and observability from day one. • Fixing what already hurts: slow endpoints, flaky jobs, integrations that drift out of sync. I use AI in my own engineering too: Claude skills for development and data analysis, an investigation skill that gathers evidence across sources to find a bug's root cause, and a classifier that triages marketplace job posts against a weighted rubric. How it works: 1. Send me the problem. Within one business day you get my questions or a short plan. 2. For anything non-trivial, we start with a paid discovery: a design doc or technical assessment with scope, risks and a fixed price. 3. Delivery in fixed-price milestones or hourly, with a written update after every working session. You own the code. 4. After launch: hourly support or a monthly retainer. You work directly with me, the engineer writing the code. Tell me what you're building or what's breaking. If I'm not the right fit, I'll say so.