Rafael – LLM, Python, AI agent development

Enterprise AI platform for healthcare-focused knowledge workflows, built around reliable agent orchestration, model evaluation, observability, and production deployment across cloud and Databricks environments.

• evolved AI agent systems using advanced frameworks, custom orchestration layers, reliable tool use, iterative reasoning loops, and multi-agent collaboration;
• enhanced LLMOps observability and evaluation pipelines with automated quality checks, tracing, metrics, and monitoring;
• designed cloud-agnostic training, inference, CI/CD, and agent execution pipelines across AWS, GCP, Azure, and Databricks;
• collaborated with data scientists and data engineers on scalable data pipelines, vectorization, experiment tracking, and model/agent serving.

Production-ready RAG chatbot with a Telegram interface, designed to answer questions from custom knowledge sources with low operational cost and clear observability.

• built an end-to-end RAG pipeline using PostgreSQL with pgvector for semantic retrieval;
• implemented multi-agent orchestration for query processing, retrieval, and response generation;
• integrated OpenAI and Groq for flexible LLM inference;
• packaged the system with Docker, Google Cloud Run deployment config, Alembic migrations, and Langfuse tracing/prompt versioning.

Data science and machine learning work for scalable product and business decision workflows, combining analysis, dashboards, experimentation, and model development.

• conducted data analysis and built dashboards to support business decisions;
• built and deployed scalable machine learning models for high-impact business problems;
• designed and analyzed A/B tests to optimize product features and strategies;
• monitored model performance and documented processes, findings, and reproducible workflows for cross-functional stakeholders.

Sports analytics computer vision platform for high-volume video/image inference, model training, deployment, and operational dashboards in a cloud environment.

• developed end-to-end computer vision pipelines from training to production inference;
• deployed vision models on GCP using GCE, Cloud Run, GKE, and Vertex AI Pipelines;
• optimized inference workflows to process large volumes of data in seconds;
• trained models with PyTorch, TensorFlow, and YOLO, and improved annotation workflows with Roboflow;
• managed team priorities and cloud costs while delivering client-facing analytics capabilities.