Rafael COELHO

/ CV

Rafael Coelho

Senior AI Engineer · Agentic AI, LLMOps & MLOps · Brazil

Senior AI Engineer with 6+ years designing, deploying, and operating production agentic and ML systems on Kubernetes — not just models, but the platforms, pipelines, and observability that make them reliable at scale. Delivered government-scale MLOps platforms (1.4M+ citizens served, 7.7M+ production interactions) and Kubernetes-native multi-agent AI systems running on 100% free-tier inference. Hands-on across the full lifecycle — data engineering, model training and serving, IaC, container orchestration, agentic orchestration with LangGraph, and observability at scale.

GitHub ↗

Areas of expertise

Data Science Data Analytics Machine Learning Deep Learning Generative AI Large Language Models (LLMs) Agentic AI Computer Vision MLOps DevOps Data Engineering Automation & Orchestration Cloud Platforms Databases Cybersecurity Pentesting Open Source Intelligence (OSINT)

Highlights

  • → Built a Real-Time MLOps platform on Kubernetes — 3 concurrent pipelines (fraud detection, delivery-time prediction, customer segmentation) fusing River ML incremental learning with CatBoost/scikit-learn batch models on Kafka, Spark, Delta Lake; 39 FastAPI endpoints, full GitOps (Terraform / Helm / ArgoCD / GitLab CI), 50+ Prometheus metrics across 11 Grafana dashboards.
  • → Architected COELHO Nexus — a Kubernetes-native platform running three production AI features (FastAPI, FastHTML, FastMCP, LangGraph, DeepAgents, Neo4j, Qdrant) with evaluation-gated pipelines and full LangFuse/OpenTelemetry observability, on 100% free-tier inference via a self-learning bandit router.
  • → Delivered production ML platforms at government scale for state government clients (1.4M+ citizens served, 7.7M+ interactions processed) — recommendation systems, semantic search, and multi-cloud speech AI, each on Kubernetes with full observability and CI/CD.
  • → Domain experience across Public Sector/GovTech, Logistics, Financial Markets, Real Estate, Computer Vision, and Cybersecurity.

Experience

Independent AI Engineer Self-directed · February 2026 – Present
  • → COELHO Nexus: Kubernetes-native platform, 3 production AI apps (doc distillation, YouTube GraphRAG, research digests), 124 REST endpoints, 18 Terraform modules, self-learning bandit LLM router with evaluation-gated pipelines, full LangFuse/OpenTelemetry/Grafana LGTM observability, $0 inference cost.
  • → COELHO RealTime: Real-time MLOps platform on Kubernetes running 3 concurrent pipelines (fraud detection, delivery-time prediction, customer segmentation), combining River ML incremental learning with CatBoost/scikit-learn batch models on Kafka (KRaft)/Spark/Delta Lake; 39 FastAPI endpoints across 3 versioned routers, full GitOps delivery, 50+ Prometheus metrics across 11 Grafana dashboards.
Senior AI Developer GovTech sector (client confidential) · April 2025 – April 2026
  • → Four production ML/MLOps platforms for Brazilian state government clients, each on Kubernetes with full observability and CI/CD.
  • → Citizen Recommendation Platform (state government client): Recommendation engine for 1.4M+ citizens, 7.7M+ interactions, hourly Delta Lake ETL; ALS models served at <100ms via FastAPI.
  • → Employment Matching Platform (state government client): Semantic job-matching pipeline (Portuguese-BERT, Qdrant) over public job listings, closing the loop with automatic citizen notifications.
  • → Speech-to-Speech: Multi-tenant, billable speech API (STT/TTS/conversational voice AI) abstracting AWS/Google Cloud/OpenAI as interchangeable providers; load-tested for p50/p95/p99 latency and concurrency.
  • → Speech-to-Speech (GPU prototype): Self-hosted Whisper + Kokoro via NVIDIA Triton; pivoted to the managed multi-cloud architecture above after GPU quantization still fell short of latency targets at scale.
Independent AI Engineer Self-directed · October 2023 – March 2025
  • → Built and shipped seven open-source AI/Data Science platforms as a personal portfolio: COELHO Finance, COELHO VISION, Formula 1 Analytics, Pentesting Automation, COELHO GenAI, COELHO Agents, and YouTube Content Search — see Selected Work for details.
Machine Learning Engineer (Contract) PropTech company (name confidential) · October 2024 – February 2025
  • → Designed recommendation strategies for real-estate property discovery; built an embedding-based similarity search pipeline (cosine similarity, Qdrant vector DB) with a scalable indexing strategy for fast, accurate property lookups.
  • → Engineered LangChain autonomous agents automating multi-step business workflows, integrated with existing systems in collaboration with cross-functional teams.
Data Science Consultant Global Big Four consulting firm (name confidential) · June 2022 – September 2023
  • → Delivered DS/AI and market-intelligence consulting across electric-utility, telecom, and sugar-ethanol clients; authored market research and competitive-landscape deliverables presented to client management.
  • → Built multi-source data pipelines (Azure Data Factory, Logic Apps, Polars/Pandas/PySpark) and interactive Dash/Plotly dashboards for market sizing and KPIs.
Data Science Intern E-commerce & retail company (name confidential) · June 2021 – January 2022
  • → Built predictive ML/DL models (scikit-learn, TensorFlow) from web-scraped and internal data; shipped analysis as internal Streamlit data-science web apps.
Business Intelligence Intern E-commerce & retail company (name confidential) · January 2020 – June 2021
  • → Consolidated logistics data into BI dashboards (Tableau, Looker); led SQL-based data structuring and analysis for internal stakeholders.

Education

Bachelor of Science in Mathematics
Federal University of Paraná (UFPR) · graduated August 2024

Languages

  • Portuguese · Native
  • English · Professional

Interests

Artificial Intelligence AI Agents Generative AI Cybersecurity Computer Vision Quantum Computing Physics Financial Markets