Stefano Meschiari, Ph.D.
Data Science Technical Lead, Duo Security
I am an experienced data scientist with a background in scientific research. I have worked on projects that involved researching new algorithms and approaches to attacking hard problems, analyzing complex, label-poor datasets, building novel production ETL and ML platforms, helping to bring intelligent capabilities to fruition for users, and creating functional data tooling.
I enjoy working with cross-functional teams to create, polish, and evolve from napkin ideas to production. In particular, I love collaborating with product and design teams to coherently integrate data science into their vision, and helping them test out new concepts and visualizations with data science-driven prototypes and customer discovery.
Senior Data Scientist
Aug 2017-Feb 2019
I lead the research and technical development of the data science platform that powers the Duo Trust Monitor product feature.
My work at Duo includes:
- Develop pipeline components and infrastructure that analyze data, build models, and surface possible threats and authentication anomalies at scale, using Spark, SparkML, and H2O.
- Research foundational supervised and unsupervised algorithms tailored to the security domain, with a particular attention to simplicity, explainability, and scalability. Translate internal domain expertise into expert rule layers and heuristics. Analyze our vast authentication dataset to mine new patterns of suspicious behavior.
- Collaborate with the product and design teams to understand how to shape algorithms and visualizations to address with our customer needs, simplify their operations, and remove friction. Work with customers via interviews, observations, and testing advanced development of new capabilities.
Product Data Scientist
Feb. 2016-Jun. 2017
- Created and improved on prototype machine learning tools and pipelines to model student outcomes.
- Prototyped new product ideas and internal tooling that employ machine learning, novel summary statistics and visualizations.
- Maintained and improved the custom modeling platform. Reduced training and scoring running time (and cloud costs) by half.
- Independently developed components for end-to-end Data Science projects:
- Back-end (Node.JS/Express, created new APIs and services exposing new functionality to internal services)
- Front-end (React, HighCharts, and custom-built components and visualizations).
W. J. McDonald Postdoctoral Fellow
SAVE/Point, Principal Investigator
- Led the data analysis effort for the Lick-Carnegie science collaboration (~20 scientists across the United States). Analyzed time series data using my Markov-Chain Monte Carlo code, Systemic. Systemic has been used to discover more than 40 new planetary systems.
- Wrote high-performance, parallelized codes to solve ordinary and partial differential equations modeling planet formation.
- Developed Super Planet Crash, an HTML5/JS game that was played more than 15 million times and was covered by The Verge, IO9, Huffington Post, and others; and Systemic Live, an HTML5/JS web app used at Caltech, UF, UT, MIT, SJSU, UD, Yale, Columbia, Coursera MOOC to teach students about data analysis and modeling.
- Machine Learning: Building supervised & unsupervised classification and regression pipelines via state of the art and custom algorithms; devising high-performance statistical and numerical methods; time series analysis and forecasting; architecting high-volume ETL and machine learning pipelines.
- Soft skills: Cross-team collaboration, project management, and leadership; mentoring and advising.
- 2012 – Doctor of Philosophy (Astronomy & Astrophysics), University of California at Santa Cruz. Received Whitford Prize for highest achievement in research, coursework, and teaching.
- 2006 – Master of Science (Astronomy, with highest honors), University of Bologna
- 2004 – Bachelor of Science (Astronomy, with highest honors), University of Bologna
Publications, Academic Honors and Awards
- Published 8 first-author publications on time series analysis, numerical optimization, and Monte-Carlo simulations (cited 224 times); a total of 17 refereed papers (cited 992 times). See research page.
- 2014 – Meschiari, S. (PI), Ludwig, R., Green, J., Interactive Education Tools in the Public Square (Award: $2,800, for creating an interactive outreach experience); Bringing the Tools of Research Direct to the UT Classroom: Systemic, a Virtual Lab for Students (Award: $87,710)
- 2010 – Award for Excellence in Teaching
- 2008 – Whitford Prize for graduate academic performance
- 2006 – Regents’ Fellow, University of California