Summary
+8 years building production ML systems: ranking, LTV, experimentation, and LLM applications across subscriptions, marketplaces, and logistics
Spent the last eight years convincing computers to make decent predictions and executives that nobody can predict next week's lottery numbers. Expert at explaining why everyone's favorite metric is wrong and why that 99.9% accuracy model definitely won't work in production. Fluent in the art of setting realistic expectations and building pipelines that don't wake the team at 3am. Have mastered the dark art of productive async collaboration across time zones without losing sanity.
Experience
Senior Data Scientist · RevenueCat · Apr 2025 - present
- Built a per-user intent model (cross-app subscription data, learned embeddings for high-cardinality features) that routes users to optimized checkout flows. Top-decile users convert at 1.46x the baseline rate.
- Shipped an end-to-end predicted LTV feature (survival models, training pipelines, backend, frontend) to let developers pick experiment winners by long-term revenue. Improved winner prediction accuracy by +12pp.
- Quantified ~$960K/year in unrealized revenue via counterfactual simulation across 11K+ experiments. Designed a Thompson Sampling system for dynamic traffic allocation using inverse probability weighting.
- Validated that granular engagement signals predict a 5x lift in paid conversion rate. Architected a real-time behavioral event pipeline to start collecting them at scale.
- Owned multiple ML initiatives end-to-end as the sole data scientist on a cross-functional team, from feasibility analysis and product definition to cross-team coordination and production deployment.
Senior Data Scientist · Wallapop · Sep 2023 - Mar 2025
- Led ML initiatives for the search team focusing on matching, ranking, and software best practices. Used Solr as the search engine.
- Trained a ranker ML model and deployed it to Solr, improving search-to-transaction ratio by +1.5%, adding 600K€ annually.
- Trained and deployed a BERT model for a query classification service. Improved the search-to-transaction ratio by 1%, adding 400K€ annually.
- Trained a PoC ranker model that used real-time features such as item popularity. Improved offline NDCG metrics by +10%.
- Developed PoC solutions for query understanding (intent extraction from queries and structured attribute extraction from descriptions) using LLMs.
- Refactored ETLs from manually executed notebooks to Spark jobs. Reduced execution time from days to hours, improving developer experience and scalability.
- Organized events to increase machine learning visibility: internal hackathons, Meetups, and conferences.
Senior Data Scientist · Stuart · Nov 2019 - Sep 2023
- Deployed a service that improved ETA accuracy by +30% using a deep learning model. Achieved a +28% improvement in cold-start locations.
- Designed and developed pipelines to automatically train, evaluate, and deploy ETA models.
- Built a distributed pipeline to process daily all the events dumped from Kafka to S3, allowing DS to analyze and train models on it.
- Designed an experimental dispatcher engine to solve the assignment problem using Python and OR-Tools.
- Mentored a senior software engineer who wanted to specialize in machine learning and data science.
Additional Data Science Experience
- 21 Buttons (Jun 2019-Oct 2019): Built a recommender system with implicit data, and an image + text-based clothing classifier.
- Privalia (Veepee) (May 2018-Jun 2019): Built a forecasting model for clearance sales. Created a pricing engine on top of the model.
- Gauss&Neumann (Oct 2017-Feb 2018): Developed tools for monitoring and optimizing SEM campaigns using Google AdWords and Python.
Talks, writing, and projects
Education
- MSc. Physics for Complex Systems · CSIC-IFISC 2016-2017
- BSc. Physics · UAB 2012-2016