Pedro Satorre-Mulet

▪ I got selected onto Twelve Football’s 1st edition of the course (AI) “Context Engineering for Football” (a very limited cohort of 45 people worldwide was hand-picked by Founder David Sumpter himself + his comittee) - implemented a hands-on project as part of the course, to develop an AI Analyst agent; ▪ I implemented a conversational CBs & CB-Pairings AI Analyst agent: grounding an LLM in structured football event data through context engineering, RAG, and tool-use techniques. It is then delivered as a Streamlit chat front-end app, answering natural-language football scouting and analyst questions with professional-analyst-grade player comparisons & pairing recommendations. ▪ AI Analyst Agents for other positions on the pitch are on their way...

▪ Designed and built a new Sports AI & Analytics business line from scratch – single-handedly and as a parallel initiative alongside daily delivery – after proactively pitching the idea & business opportunity to the entire department, securing sponsorship and direct supervision from a department-linked share-holding Partner. ▪ Delivered end-to-end Machine Learning (ML) and Deep Learning (DL) solutions for clients: data preparation and feature engineering, model training and evaluation, experiment tracking, MLOps, deployment and monitoring. ▪ Developed Generative (GenAI) & Agentic AI solutions for clients: LLM-based knowledge assistants, retrieval-augmented (RAGs), and tool-using agents. ▪ Led a team of three technical specialists: task planning, technical guidance, code review, and delivery quality. ▪ Built decision-ready dashboards and visual analytics: translating model output into recommendations for business stakeholders. ▪ Ran cross-functional collaboration with technical and non-technical client teams, from requirements workshops through to hand-over.

▪ State-of-the-art and cutting-edge MSc. Data Science thesis’ research project – modelling football shots as heterogeneous graphs of all players, the ball, as well as the instantaneous spatial & physical context, and then applying Heterogeneous Graph Transformers (HGTs) to classify shot outcomes and estimate their respective goal probability (i.e. xG = Expected Goals); ▪ Successfully demonstrated statistically that HGTs significantly improve the shot outcome classification, and their respective goal-probability estimation & calibration over traditional xG models. This thesis was directly supervised under Dr. Gabriel Anzer (Head of Football Data Analytics at RB Leipzig). ▪ In addition, this project was selected by Founder David Sumpter as 1 of 4 finalists for Twelve Football’s “Pitch to the Pros #3”.