InstaDeep  -  Research Scientist

• Designed and implemented different multi-modal LLMs for multiple genomics applications including gene expression analysis, single-cell transcriptomics, annotation, and alignment
• Refactored existing training pipelines to adapt to different needs such as multitask training or text-based prediction
• Contributed to the development of an internal Transformers library
• Designed and supervised multiple hiring pipelines for different research positions at the company

Google Brain (now Google DeepMind) -  Research Scientist Intern

• Designed and implemented a multi-objective meta evolution framework, MetaPG, to simultaneously optimize multiple Reinforcement Learning goals and understand underlying tradeoffs
• When applied to a use case on evolving RL algorithms that optimize high returns, generalizability, and stability, our method was able to find algorithms that improve upon SAC's performance and generalizability by 3% and 17%, respectively, and reduce instability up to 65%
• Actively engaged with researchers among the Reinforcement Learning, AutoML, and Robotics teams at Google Brain

MIT  -  Graduate Research Assistant

• Conducted fully-funded research on Reinforcement Learning and Mathematical Programming algorithms for Dynamic Resource Management in satellite mega-constellations
• Initiated and led a sponsored 2-year long research project on RL for drug discovery
• Led a team of 4 graduate students working on the design of the next generation of Resource Allocation and Revenue Management systems for highly-flexible mega-constellations
• Responsible for coordinating monthly reports to a sponsor company (SES), holding a semiannual review with the CTO and other senior managers, and co-organizing periodic knowledge transfer workshops with engineers at the company
• Mentored 10 undergraduate students and directly supervised 4 undergraduate Thesis

Novartis AI Innovation Lab  -  Machine Learning Engineer Intern

• Developed Deep Reinforcement Learning models for molecular design and molecular property optimization, achieving better performance than state-of-the-art models on a subset of tests from the Guacamol benchmark suite
• The project was framed inside a collaboration between Novartis and Microsoft Research, I routinely presented my work to researchers in both companies

ARCVI  -  Data Scientist

• Developed injury risk prediction models that outperformed previous methods' accuracy by 20% (impact of 90-170k€)
• Defined a new disposal architecture for 1,500 vending machines by means of Mathematical Programming, with an estimated impact of 380k€
• Developed consumption likelihood models for retail vendors and designed commercial strategies using behavioral economics principles
• Co-authored the business report Adoption and Impact of Big Data and Advanced Analytics in Spain [Spanish version]

Barcelona Supercomputing Center  -  Undergraduate Research Assistant

• Adapted different Machine Learning models for large-scale multiprocessing architectures
• After testing on the MareNostrum 4 supercomputer, our algorithms resulted in a 10x speed increase for high-dimensional clustering tasks