Chrome Data Scientist

I will be taking a career break for 4 months between roles. During this time, I will be: > completing several large self supported bike tours, specifically one from Michigan to Georgia (http://www.innocentheroine.com/2022/09/the-journey-of-1000-thank-yous-michigan.html) and another likely in Michigan> pursuing advisement and consulting opportunities on an adhoc basis (have a few planned at the moment around building out a successful ML/AI org). If you are interested in opening a project with me, please contact on LinkedIn. > writing, both technical and non technical works > Working as an ice cream scooper (the Ish Creamery) and bike mechanic (Einstein Cycles, Traverse City) > Traveling and enjoying my favorite season, Fall.

Afresh provides optimizations and reduction in food waste in Fresh departments by leveraging available data from the grocery stores and building a demand forecaster and inventory estimator. I led the Machine Learning Ops, Applied Science, Data Science, and Analytics team at Afresh (org size: 14). My team covered productionalization and deployment of the forecaster to over 1000 stores, algorithmic enhancements (from probabilistic modeling to ML, specifically LSTM development), research leveraging insights from data, and business analytics. I grew the team from 6 to 14 during my 7 month tenure and set structure to build and scale the team further.

Supply Chain Products builds and supports tooling that underpins how Facebook operates as a business. Our product area is split into two major points of focus - data center supply chain and purchasing ecosystem. The data center side of the team builds functionality to operate our warehouses, orchestrate repairs, move relevant parts to where they need to be, and how we decommission servers that contained user information. The purchasing ecosystem builds to support the invoicing, purchasing, supplier management, contracts, and sourcing that Facebook relies on in order to function as a business. We optimize for automating and scaling processes with our tooling while also creating a frictionless user experience. Our organization currently has 10 data scientists and has had > 100% growth across both teams in the last year. We value diversity highly in our team, products, and company.

Initially I worked on Messenger Core Experience before switching to Enterprise Products in March 2020. My role included advanced modeling, goaling, metrics, experimentation, building dashboards, and ensuring product success. Specifically, I was the Data Science lead for our text/sticker suggestion product, Infra Systems, and Infra Quality. On Enterprise Products, I initially started as the first Data Scientist on Supply Chain Products, supporting 150+ engineers as we built out the team and defined the Data Science Function. Notable projects I led were:* Social Impact as a Goal type across all of Enterprise Products (600+ org) * Goaling and Roadmap Framework + Standards for Supply Chain Products (150+ org) * Multiple research projects, including quantifying accuracy on our Invoice Classifier (determining accuracy at scale for an AI labeler with unlabeled data) and determining opportunities for Facebook to accelerate their Diverse Spend initiative

I am part of the Outlook Mail App telemetry team and my role was to engender the data science portion of this team, which included:NLP- Led development on an NLP engine "Text Miner" that extracted relevant noun/adj pairs from store reviews and surfaced them to our project management team in a meaningful fashion. - Initialized Text Miner with several other Outlook endpoints and in Office Customer Voice. - Managed a team of 3 interns to expand Text Miner to surface info from Tweets about Outlook Mail Machine Learning- Built a Random Forest Classifier to predict retention in first time Outlook users, led to the discovery that the major cause of user loss was associate with that user's crash rate Analytics and A/B Testing - Since the team had no analyst, I often was on call for putting out data fires, such as a slow loss of Monthly Active Users due to GDPR and Mail unpinning. My assistance allowed our PM team to work faster. - Assisted with A/B testing, often being brought in when trying to statistically vet questionable results (i.e needed to perform a Chi-Square test) - Multiple long term analysis reports exploring Retention and the impact of released features. Teaching- About once a month led a "Lunch and Learn" for the other data scientists in Outlook on other endpoints- Established a "Data Science Research Pipeline" in Outlook - Wrote much of the documentation for Pyscope internally

As part of a growing small company (I was employee #98!), I led the development on predictive ad modeling and worked extensively on data quality. My accomplishments were:- Built a Random Forest Regressor that predicted audience retention when certain ads were aired together in a particular sequence. - Built an anomaly detector for our scheduling team to adaptively identify when ads had changed compared to our directory (previously was hard coded and prone to error)- Identified many ad watcher behaviors through research and analytics. These included noting that effective ad sequencing based on demographic and show profile could increase ad viewership upwards of 50%, identifying the effectiveness of short versus long ads, realizing that ad length potency depending highly on demographic and show type (i.e. prime time or day time). These were all added to the predictive model for dramatic accuracy gains. - Made strides in data quality and initiated several engineering overhauls after data investigations. - Assisted building several metrics for determining ad success - Expanding the number of female engineers/scientists at iSpot.tv during my tenure

* Developed a program to assimilate, analyze, and predict user Daily Activity using weather information and Fitness Tracker Data (Jawbone, Endomondo)* Wrote APIs using Selenium to web scrape data from Weather Underground.* Data managed through SQL, Pandas and Scipy were used for analysis, and Matplotlib and Seaborn were used for visualizations. A combination of classification by day of week, Sckit-learn nearest-neighbor machine learning, and weighted summing used to predict user activity.* Built an app using Flask and VPS dime which is hosted at http://loisks.xyz

* Thesis on 1-10 eV Low Energy Ion Data Analysis using the Van Allen Probes HOPE, EMFISIS, and EFW instruments. Expected Defense: August 2016. * Explored low energy ion depletion in the high energy tail of the plasmasphere in conjunction with plasma wave activity* Large statistical analysis projects, developed several analytical models in python to explore plasma wave resonance, and wrote an algorithm to classify pitch angle distributions over several years of data* Also analyzed spacecraft surface charging trends in Van Allen Probes* Used numpy, pandas, scipy, matplotlib, f2py, and spacepy in Python in addition to Fortran 77/90, IDL, and Matlab during project

* Worked with incoherent and coherent scatter radar data (SuperDARN and EISCAT).* modified software to do specific ray tracing for our events* also performed large statistical analysis with 1.5 Tb of data to detemine best location of ground scatter at the SANAE radar station in Anctartica * Used Python Pandas, Numpy, SQL, Matplotlib, and Scipy to perform analysis

* Assisted in building a RANS test bed in Python/openFOAM for jets* Final Project available here: http://compphysworkshop.lanl.gov/2015_FinalReports_LAUR.pdf

* Part of NSF REU project, worked with Tiago Ribiero on using a combination of IRAF and Python to analyze infrared observational data from SOAR of cataclysmic variables

* Used the CIPS Imager as part of the AIM satellite mission to determine ice water content in polar mesospheric clouds* compared results with an analytic mode and the NOGAPS-ALPHA model * Project results in an undergraduate thesis and several presentations, including an AGU poster * Used primarily IDL in this project

* Worked in an experimental lab synthesizing gold nanorods and developing a setup to apply AC voltage to test for their alignment* Also wrote software to determine degree of alignment of colloidal particles in a sample* Resulted in a publication and several presentations