Paul George Email & Phone Number

San Francisco, California, Us

Toronto, Ontario, Ca

Perpetua is an eCommence platform that helps clients optimize digital marketing campaigns across multiple disparate marketplaces. I joined Perpetua to help improve scalability, developer productivity, and Perpetua's ability to manage advertising campaigns with increasing sophistication through the application of established best-practices and technical leadership. My role was very technical and a large portion of my time was dedicated to rapid prototyping and mentorship. I lead efforts to improve data quality and application performance, introduce new infrastructure and best practices, and to simplify existing codebases thereby reducing technical debt. I also helped introduce a more rigorous documentation and RFC process as well as Weekly Activity Reports. I joined Perpetua in February 2022 as Staff Engineer and was promoted to Director of Engineering in October 2023.+ Overhauled main Airflow DAG and reduced runtime from 36 hours to <5 hours through incremental data loading, fast ingestion into Postgres, and general code improvements.+ Lead implementation of medallion data architecture with introduction of new fast, high quality “silver” datasets.+ Improved analytics reporting through various SQL query optimizations and introduction of a new analytics service based on FastAPI.+ Lead major effort to overhaul data for core bidding engine, including architecting and implementing feature ETL DAGs and online serving using BigQuery StorageAPI. Achieved 100X improvement in throughput, resource utilization, while concurrently improving data quality and auditability.+ Prototyped next-generation bid optimization and model development with serverless PySpark on Dataproc.+ Languages/Technologies: Python, Postgres, Docker, GCP (CloudSQL, BigQuery, Composer, Dataproc), Terraform+ Libraries/Frameworks: Pandas, Numpy, PySpark, FastAPI, Django/Django REST/Django Ninja

Toronto, Ontario, Ca

San Francisco, California, Us

I was a member of the Valuation Team at Opendoor and primarily worked on improving the accuracy, performance, and scalability of Opendoor’s Valuation Models (OVM) for predicting the fair-market valuation of homes.+ Analyzed and optimized feature fetching queries (PostgreSQL) for real-time valuation serving.+ Redesigned and implemented model training and backtest Airflow DAGs to reduce runtime and dramatically reduce AWS costs.+ Optimized neural network valuation model preprocessing code (Pandas, Numpy) to reduce real-time prediction time from 40s to sub-second.+ Implemented multi-GPU parallel neural network model training in FastAI to significantly reduce training time and costs.+ Helped migrate neural network valuation models from legacy FastAI-based codebase to PyTorch to improve performance, reduce technical debt, and improve research flows.+ Developed and maintained various data and model debugging tools and methodologies, including Jupyter notebooks and monitoring scripts to perform continuous model monitoring.+ Languages/Technologies: Python, Docker, Postgresql, AWS (ECS, RDS, S3, ASG, Athena, etc.)+ Libraries/Frameworks: Pandas, Numpy, Scikit Learn, PyTorch

The Meson Capital "Gravity" fund is a machine-learning based value fund focused on algorithmically identifying equities to sell short. I joined the team as the lead engineer unofficially in November 2016 and then officially in November 2017. I built core data and machine learning infrastructure as well as Meson's proprietary strategy backtesting platform. The infrastructure was written primarily in Python, relied heavily upon Pandas, Numpy, and Scikit Learn for ML models, and was deployed both on "in-house" hardware as well as on AWS. I also contributed to machine learning feature generation, model development, and research. Gravity strategies utilized a custom-built time-series of thousands of fundamental and price features from Capital IQ's Xpressfeed database spanning 20+ distinct datasets to identify out-performing (long) and under-performing (short) equities.+ Lead engineer+ Developed and maintained backend infrastructure for transforming and ingesting multiple terabytes of fundamental financial (10K/10Q) and price features from Capital IQ Xpressfeed.+ Developed and maintained machine learning API for training thousands of Scikit Learn models in parallel on clusters AWS ECS.+ Developed and maintained Meson Capital Gravity's proprietary portfolio and strategy backtesting framework. Included the effects of liquidity, borrow availability, and interest for more accurate short-selling simulations.+ Built "walk-forward" model training and prediction for causal training on time series.+ Extensively studied model performance under various features sets and groupings (e.g. company sector, etc.) using both standard and custom scoring methodologies.+ Languages/Technologies: Python, Docker, AWS (ECS, RDS, S3, ASG, Athena, etc.), MSSQL 2017 on Linux+ Libraries/Frameworks: Pandas, Numpy, Scikit Learn, Django/Django REST

Videonote.com is an education platform that enables content creators, such as university professors, to upload and annotate videos with text and multimedia. Users can quickly search video libraries for specific topics and review specific sections based on content, difficulty, or popularity.I co-founded Videonote in 2008 while finishing up my PhD at Cornell University; the platform has been used for ~20 classes every academic semester from 2008 until present. I maintain an active interest in improving and developing Videonote and recently lead an outsourced development team through a complete site overhaul. We migrated from monolithic ASP.NET application to decoupled single-page, responsive AngularJS application and a Django API.+ Co-founder and engineer+ Manage outsource developer team: plan daily updates, review commits, file issues.+ Develop new features specifications and provide mock-up implementation.+ Languages/Technologies: Python, AngularJS, PostgreSQL, SCSS, Docker, AWS

Santa Clara, California, Us

SolveBio enables precision, genome-based medicine by providing genetics labs and biopharma with the ability to quickly analyze and interpret information in complex, disparate biological datasets.In many ways, the technical problems I worked on at SolveBio were similar to those at Palantir: providing interfaces o enable domain experts (in this case, clinical geneticists, researchers, etc.) to quickly access and analyze dozens of disparate datasets. My primary responsibility was to design and build a flexible data ingestion pipeline for transforming and importing a variety of datasets (SQL, CSV, Excel, XML, etc.) into our Elasticsearch database. Additionally, I worked closely with geneticists to build a suite of genomic webservice APIs, including a proprietary and configurable real-time variant classification engine and predictor based on ENSEMBL's VEP.+ Co-Founder and engineer+ Developed and maintained backend infrastructure for transforming and importing disparate genomic datasets into SolveBio's dataset library.+ Developer a genetic variation classifier for near real-time analysis of variations based on a configurable rule set.+ Co-developed infrastructure for fast annotation of million row VCF files.+ Exposed most backend services through JSON API (Django, Django REST Framework).+ Languages/Technologies: Python, Hadoop/MR, Elasticsearch, SQL

Denver, Colorado, Us

+ Client-facing software engineer with specific focus on developing the Palantir Government and Finance applications for use in the commercial business sector.+ Co-developed solutions for enabling real-time analysis of large datasets in multiple business spaces, including fraud and cyber threat investigations, mortgage portfolio optimization and customer segmentation.+ Languages/Technologies: Java, SQL, Rails

Doctor Of Philosophy (Phd), Electrical And Computer Engineering

Master Of Engineering (Meng), Electrical And Computer Engineering

Bs, Electrical And Computer Engineering