Daniel Berry Email & Phone Number

Corvallis, Or, Us

San Francisco, Ca, Us

Experimentation Science Lead - working to scale the velocity and impact of the experimentation program. Achievements include:* Helped scale experimentation volume over 5x (from <50/year to >250/year) which led to >10 experiments with strong, positive impact on company north star metric (and many more experiments that drove guardrails or departmental KPIs)* Implemented state-of-the-art statistical methodologies to increase program efficiency and effectiveness. These include covariate adjustment through linear models, retrospective A/A tests, and anytime-valid confidence sequences. * Drove the development and adoption of long-term holdbacks to measure the cumulative impact of experimentation teams.I am responsible for driving the sophistication and effectiveness of the experimentation program. I own statistical validity and reliability of experiment results, tools to automate experiment planning and analysis, and training and cultural programs to increase experimentation knowledge and capabilities across the company.

Boston, Ma, Us

As a lead data scientist, I ran analytical initiatives to support existing products or help develop new ones, including:- Designed and built an A/B testing framework, whose experiments led to 20% lower distracted driving frequency and a resulting estimated minimum savings of $1.5 million in loss expenses over 5 years.- Served as technical product owner for launch of new pricing model.- Built inventory forecasting models to support supply chain team, resulting in >$20MM in spend on program hardware.- Trained and coached junior data scientists and served as a subject matter expert. Coached junior team members in an internal modeling competition, oversaw a new analyst in implementing program health dashboards, and organized a data science meetup to share skills and techniques.

Boston, Ma, Us

Built and deployed price optimization, shipment ranking, shipment matching, and pallet optimization algorithms resulting in increased margins. Built streaming data pipelines powering real-time dashboards. Conducted simulation studies and market sizing analyses to guide business strategy. Rapidly adapted to changing business priorities and collaborated with diverse stakeholders.

Northbrook, Il, Us

Held a variety of responsibilities enabling the delivery of a telematics-based risk model for a novel usage-based-insurance product, including:- Architected and built our big data pipeline which rapidly delivers modeling datasets derived from hundreds of billions of raw telemetry events. Implemented complex feature extraction algorithms at scale, turning semi-structured user-behavior data such as GPS and accelerometer readings into interpretable features for our models. As part of this work, I led the development of software package providing a high-level API on top of Apache Spark which enabled teams to share pieces of their pipelines. We've used this to set up shared code libraries between 5 different teams and enabled data scientists to continuously integrate upstream improvements from data engineers. I used a CI/CD methodology to enable quick changes and improvements to our pipeline as well as to automate regression testing against upstream changes introduced by data engineering teams. On several occasions, these automated builds and tests rapidly alerted us to problems in our modeling pipeline.- Led a junior data scientist in the development of our machine learning model training pipeline which fits cross-validated and regularized models as well as automatically generating model diagnostics and interpretability for regulatory purposes.- Proposed and implemented the feature extraction for several novel features, improving the accuracy of our risk model.- Oversaw a data science intern in her project to set optimal initial rates for new customers on the rating plan, incorporating retention and profitability constraints.- Provided guidance and assistance across the department as a subject matter expert on Apache Spark.

Northbrook, Il, Us

Built a machine learning system to predict future car state for Arity, Allstate’s telematics startup. Created a data pipeline using Sqoop, Hive, and Spark tasks managed in Airflow. Fit intermediate GLM, GBM, and time series models in Python which were combined to form the final model. Implemented the production model in Groovy and hosted it on an internal Prediction-as-a-Service platform to serve the model via an API. Collaborated with app engineers to design logging infrastructure and model performance monitoring.Prototyped a Spark pipeline to build 200,000 GBM models, one for each user, to solve an active research problem. Generated a dataset using a novel data pooling approach where negative observations were synthesized using similar data from other users. Utilized spatial data structures within the distributed job to boost performance. Pipeline featured ETL, feature extraction, model training, and model analysis using vehicle and phone sensor data. Developed Spark new hire training program, contributed to internal software packages, gave seminars on hierarchical linear models and big data visualization, and created a marketing map using 5 billion GPS points.

Blacksburg, Va, Us

Led a team of 5 masters students in providing statistical consulting services for researchers including data manipulation (data transformation and statistical programming), data visualization, and data analysis (model building and interpretation). Consulted for clients in a variety of departments including business, engineering, architecture, biomaterials, genetics, agriculture, and osteopathic medicine.

Blacksburg, Va, Us

Providing statistical collaboration services to help clients (faculty and graduate students) meet their research objectives

Blacksburg, Va, Us

Lead instructional recitation sections and graded assignments

Northbrook, Il, Us

Developed a model to detect car crashes from embedded sensor data such as accelerometers, GPS, and OBD2 ports. Created modeling file using Python, Spark, and Hive. Extracted fea- tures using Python toolkits and trained GBM models using XGBoost and scikit-learn. Demon- strated 50-fold accuracy improvement over original crash detection logic.

Northbrook, Il, Us

Prototyped a natural language feature extraction system in Python to mine accident claim notes and developed a simple GBM model to predict auto losses in R.

Average improvement of two letter grades among students meeting weekly.

Kansas City, Mo, Us

Developed internal tools in R and Python for data visualization and predictive modeling.

Baton Rouge, La, Us

Researched computational combinatorics with Dr. Valerio DeAngelis of Xavier University of Louisiana (XULA). Implemented and applied transforms to a database of integer sequences using Mathematica. Paper title: Umbral Calculus and the Boustrophedon Transform.

Master’S Degree, Statistics

Master'S Degree, Statistics

Bachelor'S Degree, Mathematics