Using RDataTracker
ER Boose, H Forest, BS Lerner, LA Perez - 2014
Software Engineer/AI Researcher
Software Engineer and Technical Leader at Waymo with 8 years experience in machine learning and 6+ years with PyTorch and Tensorflow frameworks. Expert in large-scale model training for behavior prediction, multimodal content understanding, and ranking (eg. wayformer, transformers, graph attention networks, seq2seq models).
Work Experience
[2022 - ] Waymo | Software Engineer
Software lead working across behavior prediction, planner, and simulation to improve trajectory modeling and driver realism.
[2020 - 2022] Meta AI (formerly Facebook AI) | Software Engineer
Technical lead working with 5 teams with 30+ engineers across key products (Instagram Reels, News Feed, and Marketplace)
- Reengineered Meta AI’s multimodal content-understandplatform in PyTorch, leading migration of 100+ models
- Led development of PyTorch Recipes, enabling new modelsin IG Reels (+2.3% time spent) and Marketplace (+1.3% in CTR).
- Maintained reproducible SOTA results across video, text, and audio domains with >3K daily active developers.
- Accelerated growth of organization and established technical foundations, supporting over 800+ DAUs on SOTA models.
- Published across several venues with >2.3k citations and experience across production training systems.
[2022 -] North Carolina A&T University (Meta) | Visiting Faculty
Lead instructor establishing novel algorithms for artifical intelligence and cyber-security, serving 400+ undergraduate and 20+ graduate students at the largest HBCU in the US.
- Created course (YouTube) and led the instruction to 62 students of COMP285: Algorithm Foundations forAI and Cyber-security
- Achieved a course satisfaction of 4.5/5 (+18.8%) and NPS of 73 with >10% of students accepting full-time positions at Meta.
[2018 - 2022] DeepMind (Alphabet) | Senior ResearchEngineer
Technical lead for graph neural networks research team (7+ research engineers),collaborating across product areas (GoogleMaps, YouTube, Google Cloud, Google Search) to deploy state of the art ML architectures for recommendation, ranking, and prediction.
- Productionized GNN-based model for traffic estimation in Maps (paper) reducing negative ETA outcomes in 80+ regions compared to production baseline (40+% in cities like Sydney, with average reduction of 18% globally).
- Productinized GNN-based model for video candidate generation for content filtering on YouTube recommendations, decreasing XLQ (eXtreme Low Quality) content by 20+% across key demographics (eg, young adult, lgbtq+).
[2016 - 2018] Google (Alphabet) | Senior SoftwareEngineer
Individual contributor to Procella leading support for structured queries, ETL pipelines, and map-reduce style execution.
- Developed performance-critical infrastructure for YouTube analytics platform at scale in C++. System serves 70+ million QPS and supports sub 5ms lookup queries for indexed data, enabling real-time analytics for YT creator dashboards.
[2015-2016] Okta | Software Developer
- Developed and implemented cloud service connections using Node.js to serve Charles Schwab & GE, driving $10M+ in revenue.
[2013 - 2015] Harvard University, Wyss Institute and Forest | Researcher/Developer
- Built program execution tracer for R, automated testinginfrastructure, and performance analytics framework.
- Implemented construction simulation for decentralizedAI used in the TERMEs project for real-world construction.
Skills
Languages
- Proficient: Python, C++, C
- 2+ Years Experience: Java, SQL, Javascript, R
- Hobby: Haskell, OCaml, Rust
Frameworks:
- ML: PyTorch, Tensorflow, PyTorch Lightning, Kornia
- Libraries: torchvision, torchtext, torchdata
Machine Learning/Deep Learning Techniques
- Architectures: convolutional models, graph networks, seq2seq models, transformers
- Training Tricks: optimization, distillation, ensembles
- Data: label propagation, nosy labels/data, distributional shift,
Other Technical
- Distributed systems, Map Reduce, embedding lookup, vectorized databases, SQL, sharding, storage formats
Education
[2020] Stanford University | M.S in ComputerScience - AI Specialization
- Awards: Community Impact Awards
- GPA of 3.8/4.0
- GRE of 335/340 (6/6 Writing)
[2016] Harvard University | B.A. in ComputerScience & Mathematics
- Awards : Phi Beta Kappa, Harvard Detur Book Prize, John Harvard Scholar
- Cum Laude with Honors
- GPA of 3.9/4.0
- SAT of 2390.
[2012] Nacogdoches High School
- Valedictorian
- GPA of 106.25/100
Publications
Predicting Yelp Star Reviews Based on Network Structure with Deep Learning
L Perez - arXiv preprint arXiv:1712.04350, 2017
The Effectiveness of Data Augmentation in Image Classification using Deep Learning
L Perez, J Wang - arXiv preprint arXiv:1712.04621, 2017 - arxiv.org
Using introspection to collect provenance in R
Lerner, B.; Boose, E.; Perez, L. Using Introspection to Collect Provenance in R. Informatics 2018, 5, 12. https://doi.org/10.3390/informatics5010012
The Effect of Embeddings on SQuAD v2.0
Perez, Luis. “The Effect of Embeddings on SQuAD v 2 . 0.” (2019).
Procella: Unifying serving and analytical data at YouTube
B. Chattopadhyay, P. Dutta, L. Perez, O. Tinn, A. Mccormick, A. Mokashi, P. Harvey, H. Gonzalez, D. Lomax, S. Mittal, et al. Procella: Unifying Serving and Analytical Data at YouTube. PVLDB, 12(12), 2019
Automatic Code Generation using Pre-Trained Language Models
L Perez, L Ottens, S Viswanathan - arXiv preprint arXiv:2102.10535, 2021
Mastering Terra Mystica: Applying Self-Play to Multi-agent Cooperative Board Games
L Perez - arXiv preprint arXiv:2102.10540, 2021
ETA Prediction with Graph Neural Networks in Google Maps
Austin Derrow-Pinion, Jennifer She, David Wong, Oliver Lange, Todd Hester, Luis Perez, Marc Nunkesser, Seongjae Lee, Xueying Guo, Brett Wiltshire, et al. Eta prediction with graph neural networks in google maps. CIKM, 2021.