Victor Gabriel Andrade de Melo
MLOps & AI engineer. 5+ years taking data to production, from ETL and data lakes to ML models and LLM agents. I build the boring, reliable parts: evaluation gates, observability, guardrails and deploys you can roll back.
Projects
agent-orchestrator
Multi-agent API: supervisor/worker delegation, parallel tool calls, MCP tools, guardrails and end-to-end tracing.
mcp-server-brasil
MCP server with Brazilian public data for LLM agents: CEP, CNPJ (alphanumeric too), holidays, Banco Central, IBGE.
genai-lens
Self-hosted LLM observability: traces, tokens, cost and latency, with an OpenTelemetry GenAI endpoint.
mlops-platform
MLflow end to end: lineage, gated champion/challenger promotion, drift (PSI) and hot-swapping model serving.
ml-from-scratch
Classic ML and attention in pure NumPy, every algorithm verified against scikit-learn.
Stack
Python · FastAPI · LLM agents · LangGraph · MCP · Anthropic / OpenAI / Ollama · MLflow · scikit-learn · SQL · Postgres · data lakes · Go · Docker · AWS · Azure · OpenTelemetry