Braden Poe

Senior Data Scientist

Urgently•  April 2024 - Present•  Vienna, VA (Remote)

Lead developer of end-of-network pricing model utilizing Uber’s open source H3 geospatial library, constrained optimization, and supply insights. Increased end-of-network digital acceptance by ~15% and lowered average per job costs by ~$26 equating to $1.5MM annual savings. Developed geospatial feature engineering process to enhance clustering capabilities that power market-level pricing. Enhanced business understanding of location signals by moving from ZIP/County to more granular hexagonal mappings.

Data Scientist

Urgently•  April 2022 - April 2024•  Vienna, VA (Remote)

Lead architect of tunable provider management model which algorithmically ranks towing providers based on past performance and cost metrics. Outperformed business expectations by ~10x, increasing net profit margin by ~5% per job. Utilized Airflow to create monthly refresh pipeline that updated model on latest data and implemented safeguards for drift. Created monthly Data Science Innovation meeting to uncover product gaps and brainstorm future solutions. Contributed to initial XGBoost pricing algorithm by layering location and supply-based adjustments on top of calculated market price.

Data Scientist

State Auto Insurance•  August 2020 - March 2022•  Columbus, OH

Led development on hazard classification and inspection probability products that saved Risk Engineering (RE) teams 1000's of hours annually. Wrote automated AWS pipeline that inferenced images for every new inspection in real-time and posted to RE teams’ UI. Developed custom CNNs and ResNet transfer learned models to generate > 80% accurate hazard classifiers for hard-to-classify targets such as "crack in foundation". Built an XGBoost/web-scraper ensemble to identify the failure probability for virtual home inspections. Achieved ~85% precision and controlled for long-term model drift with an active learning approach.