Nacho Martinez

Nacho Martínez is a principal data science advocate at Oracle who teaches developers to build machine-learning systems through hands-on projects. His work spans game-based prediction, agent memory, and open-source tools for exploring how language models reason.

Teaching machine learning through games

Martínez authored an Oracle and Red Bull Racing workshop that takes learners from extracting and joining racing data to training and deploying a model for real-time predictions. The workshop credits several Oracle contributors alongside him.

His League of Legends learning materials teach data extraction, model building, and neural-network basics. The accompanying optimizer uses Riot Games APIs to collect player and match data and train models that estimate a player's chance of winning. For live predictions, the project describes sending game data through a message queue to a separate process running the model, reducing the work performed on the player's computer. Its demonstration uses a smaller model to keep prediction times down, accepting less accurate results.

Building agents that remember and reason

Martínez co-teaches Agent Memory: Building Memory-Aware Agents with Richmond Alake. The course uses Oracle AI Database, LangChain, and language-model pipelines to build memory that persists across sessions. Learners create a memory manager, retrieve relevant tools through semantic search, and build pipelines that extract facts from conversations and consolidate experiences into longer-term knowledge.

His open-source Agent Reasoning project explores ways to organize a model's work before it produces an answer. Its implementations include tree of thoughts, which explores alternative reasoning paths, and self-reflection, which cycles through a draft, critique, and revision. The project also provides interfaces for comparing strategies and inspecting their intermediate behavior.