Data scientist, creative coder, and designer with 9 years of experience doing analysis, building and deploying models, and writing software in consulting and startup contexts.
Over 30 successfully completed projects with IDEO, Datascope, and other organizations
since 2017.
Clients range from Fortune 10 members to startups, across domains such as mobility, healthcare, and education.
As a data scientist at IDEO, I use human-centered design to tackle gnarly challenges in all sorts of industries.
I create data-powered software, products, and services that are rooted in an in-depth understanding of the problems facing the humans who interact with these systems.
Highlights
A redesign of the gate and boarding experience for a major airline that culminated in live testing of a novel
boarding process on 12 international and domestic flights serving 2500+ passengers in 2 airports. Built out
a front and backend system using Python and Vue in 2 weeks that enabled testing by handling passenger
data from the airline, assigning new boarding groups, and controlling the boarding process.
Created
StyleGAN MTV, which generates music videos by feeding audio spectrograms as input into
a StyleGAN generator. The Python backend uses TensorFlow and librosa for image generation, with a
Flask backend / Vue.js frontend to create a real-time interactive experience where users can play sound
from arbitrary sources and see live generated video.
In fall 2017, Datascope was acquired, giving birth to a brand new discipline at IDEO: data science.
I helped to define how data science would integrate with IDEO's creative process and demonstrated to
teammates the value of building and making with data in design projects.
Highlights
Evaluated the feasibility of predicting restaurant sales, laying the foundation for a future product. In the first two weeks I did
the engineering to create a robust end-to-end data pipeline including ingestion, cleaning/normalization,
feature engineering, and model training and evaluation. This set up the rest of the project for
success as we experimented with different features and model parameters to determine how feasible
prediction was, and what the implications for user experience might be. The project inspired
the client to continue to develop the concept with IDEO as a startup that went to market in fall 2020.
Built the backend for the Belief Checkout,
a provocative exhibit part of the Hyperhuman gallery at Munich
Creative Business Week 2018 exploring how humans might trust algorithms to make decisions that align with their values.
Users "shopped" using physical cartons representing different values such as sustainability or healthiness,
then received a receipt of groceries the algorithm purchased for them. Hyperhuman was the recipient of the
2019 Core77
Design Award for Speculative Design.
Used agent-based simulation to investigate possibilities for the design of future
autonomous vehicles and service opportunities. We explored how different vehicle parameters such as max
speed and capacity could effect differing service and logistics models.
To mimic real-world conditions at a proper level of abstraction, we built simulations in which agents
traveled street networks pulled from Open Street Maps and responded to modeled demand at actual addresses
generated by sampling from OpenAddresses and Census population datasets.
Ensured sessions at Scialog conferences are composed in a diverse way
through administering custom-built network surveys, running a simulated annealing optimization algorithm based
on survey results, and analyzing optimization outputs.
Parallelized execution of optimizer using
serverless computation with AWS Lambda, leading to more than 800x time reduction (from 2.8 days to 5 minutes).
Datascope was a data science consulting company with heavy design thinking influences. We helped our
clients understand and investigate possibilities for data they had or could collect by doing exploratory
data analysis, making prototypes, or coding production-ready software that delivered real business and human
value.
Highlights
Created an end-to-end modeling pipeline for a major paint company exploring paint color matching.
Used historical paint formula data to predict the color spectrum of a given recipe.
An optimizer used this model to find a starting point formula for humans to refine, saving both time
and expensive raw materials. Coded in Python using scikit-learn and pandas.
Created a transportation mode inference model for a mobility startup. With GPS time series data,
created map visualizations using GeoPandas and Folium to understand commutes
and get ideas for features useful to predict how someone is commuting.
Trained and evaluated different mode inference models with scikit-learn, and productionized
the mode inference model pipeline using Celery.
Skills / Tools
I am very comfortable with Python and its ecosystem of SciPy, web frameworks, and other libraries. I tend to use
Javascript when things need to be done interactively and/or on the web. While I generally try to stick
to whatever will do the job at hand best, here is a non-exhaustive, unordered list of some tools/packages I may reach for first: