Overview

Applied Scientist & Machine Learning Engineer

  • Applied Scientist, Research Engineer, specializing in multimodal AI, with a focus in computer vision (CV), and Machine Learning for Audio Applications
  • Specialized in 2D computer vision (CV) at Amazon Photos, video understanding and contextual AI at Meta Reality Labs Research (RL-R), and most recently as an applied scientist in generative audio at Soundverse, additionally leading our AI team, and ensuring engineering excellence as CTO
  • Educational background, and graduate programs in both astrophysics and fusion plasma physics

Experience

Soundverse.aiChief Technology Officer & Lead Applied Scientist
Apr 2025 – Present
  • Lead scientist, developing and training both in-house foundational models, as well as leveraging SoTA models for differentiating platform features
  • Led Audio-AI team and supervised engineering operations at Soundverse.ai
  • Designed and developed novel audio attribution feature to protect artist content from unlicensed training
Oct 2021 – Apr 2025
  • Trained large, multimodal models incorporating, image, video, audio, digital context, and natural language for personalized, contextually aware digital agents.
  • Developed RAG system, including modeling and evaluation, for end-to-end predictive interface functionality for few-shot action prediction of egocentric video for augmented reality kit
  • Developed human/machine alignment benchmarking initiative, including goal classification, prediction, and labeling efforts
  • Coordinated group of 8 engineers and researchers to deliver first-of-its-kind dataset, composed of egocentric, longitudinal video, and novel annotations via custom labeling tooling.
  • Developed multimodal, semantic search pipeline for automating high-information-yield candidates from unstructured video
Amazon PhotosApplied Scientist
Mar 2017 – Oct 2021
  • Developed CV projects for Amazon’s consumer image-management app, enabling features scaled to millions of customers:
    • Face Recognition, Object Detection, Instance Segmentation, Clustering Methods, CNN Classifiers with Semi-Supervised Training, Recommender Systems, Photo Stitching, Annotation Pipelines, Salience and Smart Cropping
  • Managed training on datasets of (and inference for) millions of customers, while preserving privacy and ensuring computational efficiency
ArenaNetData Specialist
Nov 2015 – Mar 2017
  • Commercial ML applications and general video game development, including research, design, and engineering of a recommender system to drive microtransaction sales of the online store
  • Unity prototyping for emerging AI companions
  • Courses taught include:
    • Orbital & Flight Mechanics, Rocket Propulsion, Flight Test Eng., Electricity & Magnetism, Waves & Optics
San Francisco State UniversityGraduate Research Assistant
2012 – 2013

Research under Dr. Andisheh Mahdavi

  • Independent research of analytical formation-models of heating and cooling flows of intracluster plasma from active galactic nuclei (AGN)

Education

The University of WashingtonMaster of Science in Aeronautical & Astronautical Engineering (Emphasis in Magnetic Confinement of Fusion Plasmas)
2013 – 2015
  • Focus in computational and theoretical methods in fusion plasmas
San Francisco State UniversityPost-Graduate Studies in Physics (Emphasis in Astrophysics and Cosmology)
2011 – 2013
  • Student and Research Assistant studying the formation of AGN
Northern Arizona UniversityBachelor of Science in Physics
2006 – 2011
  • Graduated magna cum laude
2008 – 2008

Skills

Machine Learning
Computer Vision
Audio AI
Deep Learning

Tools

PythonExpert
PyTorchAdvanced
SciPy stackAdvanced
numpyscikit-learnpandasetc.
Data visualizationAdvanced
plotlymatplotlibseabornstreamlitgradioetc.
Cloud ComputeProficient
AWSGCPSLURMGit
SQLIntermediate
Web DevBeginner
ReactTypescript
API GenerationBeginner
FastAPICelery

Publications

Done at Meta Reality Labs Research in colloaboration with FAIR, in this work we consider the important complementary problem of inferring that goal from multi-modal contextual observations. Solving this “goal inference” problem holds the promise of eliminating the effort needed to interact with such an agent. This work focuses on creating WAGIBench, a strong benchmark to measure progress in solving this problem using vision-language models (VLMs)

Languages

EnglishNative speaker
FrenchBeginner / Intermediate
ItalianBeginner
GermanBeginner

Awards

Theodore H. & Marie Sarchin FellowshipUW
2013
Paul A. Carlstedt Endowed FellowshipUW
2013
Honored MS graduate in the Physics and Astronomy departmentSFSU
2013
Provost Scholar of Entering Graduate ClassSFSU
2011
NAU Physics Department Chair ScholarshipNAU
2011
Bedwell Physics Award RecipientNAU
2010

Interests

Muisc Production
Logic Pro