About me

I’m a senior researcher on the PROSE team at Microsoft, where we develop state-of-the-art program synthesis technologies to make software development more accessible, productive, and fun. My current line of work focuses on using neurosymbolic techniques for applications ranging from program repair to programming by example.

Some of our recent work at PROSE includes:

• Neurosymbolic low-code formula repair to appear in OOPSLA 2022
• Neurosymbolic conditional formatting by example
• Large language model-based multi-lingual program repair

If you’re a PhD student in a related area of research (e.g. SE, PL, ML) and are interested in doing an internship with me, please reach out via email.

Before joining Microsoft, I received my PhD from MIT, under the supervision of Martin Rinard. You can find a copy of my thesis here.

I have also researched and published work on:

• using dynamic program analysis for automated machine learning (AutoML)
• generating AutoML search spaces using API docs and code corpora
• mining nearby transformations of machine learning pipelines to improve their performance
• evaluating the empirical impact of downsampling as a way to scale genetic-programming-based AutoML to large datasets
• mining data wrangling functions from collections of Python programs written to work with the same dataset
• optimizing database queries to perform dynamic imputation
• studying how developers use automatically generated bug patches
• using deep learning to improve code search using natural language
• applying active learning to infer and regenerate software
• developing a risk prediction model for pancreatic cancer using electronic health records

Before my PhD, I obtained an M.S. in Computer Science (NYU) and a B.A. in Economics (University of Pennsylvania). I have had the opportunity of working (both as an intern and full-time employee) across technology and finance companies, in a variety of roles. I’m originally from San José, Costa Rica.