Things I’ve built, researched and taught.

A Colombian Sign Language translation system reaching a 91.17% F1 score, trained on a purpose-built corpus of 1,500+ videos and deployed on Kubernetes.

A production retrieval-augmented generation platform, built with FastAPI, that lets legal teams query a corpus of 10,000+ judicial decisions.

Distributed TensorFlow training on Kubernetes with the Parameter Server strategy: TF_CONFIG topology, gRPC coordination and network policies.

End-to-end MLOps architecture on AWS: S3 + Glue ETL, SageMaker training and batch inference, Athena analytics, Step Functions orchestration and KMS encryption.

Hands-on workshop building an AI agent from scratch with LangGraph and Google Gemini: three progressive scripts, from first LLM call to a multi-node agent.

One-hour workshop training a handwritten-digit classifier with TensorFlow and MNIST: data, neural networks, the training loop and predictions.