|
|
This page contains employment and education details. For non-work projects, see the rest of the site. (Some substantial projects may be excluded due to proprietary information.)
Email: gaffney.tj@gmail.com
I worked on Omnichannel Optimization, which attempts to maximize conversions from multiple channels (online, in-app, offline).
- Developed the first prototype model for targeting omni-shoppers, delivering strong offline results that justified advancing the model to online experimentation.
- Played a key role in launching a new model to 11 alpha advertisers by ensuring campaign QA, historical data availability, system enablement, and measurement frameworks were in place. It was later rolled out to a $300M+ product.
Developed an incrementality estimation and bidding approach using experimental lift data and an ensemble of conversion models, iteratively validating and tuning via backend tests to achieve a 10X lift in estimated incremental conversions.
2022-2025- Built a custom analysis workflow for omni-shopping backend experiments, enabling statistically rigorous evaluation of offline conversion lift at scale, supporting 4-8 experiments per month, and saving significant analysis time while accelerating model improvements.
- Unblocked statistically significant experimentation for an offline model by correcting measurement issues, expanding eligible traffic (~2.5X), and designing a fallback holdout approach, ultimately enabling the first valid results and supporting general rollout.
I worked on Reddit's User Understanding team, whose main task was to features for use in models, primarily recommendations. I created specific features and established patterns for aggregating content features to users and creating user embeddings. This work focused on both batch and streaming pipelines.
Designed and implemented a collaborative-filtering–based user embedding framework with streaming updates to mitigate cold start, enabling effective use in large-scale recommendation and advertising models.
2021-2022Engineered scalable user profiling pipeline that aggregated content labels to the user level—incorporating NSFW filtering, label grouping, and temporal decay—delivered via batch (Airflow/BigQuery) and streaming (Flink) systems, with downstream user-to-subreddit mappings powered by approximate nearest neighbors.
2021-2022Built bespoke Markov chain approach to compute average time-to-discover for subreddits. Optimized and parallelized expensive matrix computation for >99% speed up.
2021-2022Built an analytics dashboard and informed serving pattern changes that increased available ad slots by ~8%.
2021-2022Identified key user-level covariates and built tooling to compute them, improving the rigor and interpretability of A/B test impact analyses.
2021-2022I worked on a team called YouTube Brand Connect, which facilitates organic ads in YouTube videos. My role was to build models for predictions and recommendations.
Designed and implemented a channel recommendation model to identify the most relevant YouTube channels for a brand based on campaign keywords and URL-derived context. The approach leveraged shared embedding spaces and novel clustering techniques to account for multimodal channel content, paired with a two-stage ranking system optimized for real-time querying at scale. This system reduced channel selection time by ~90%, reproduced expert decisions with >99% precision, and was patented.
2018-2021Applied Monte Carlo simulation techniques to estimate aggregate view outcomes for planned video lineups, addressing systematic underprediction caused by outlier effects. The approach enabled reliable percentile-based forecasting, was adopted in a patented solution, and improved planning accuracy for internal stakeholders.
2018-2021Developed a UI-driven pipeline to streamline and automate video review workflows.
2018-2021Implemented a service to support CRUD operations, exports, and reversals for a payments table.
2018-2021Worked with audience sentiment data to aggregate and analyze signals at the channel level in support of multiple backend initiatives.
2018-2021I did some side work on a project which attempted to automatically process contracts. For my part, I scraped the FCC EDGAR database to find contracts to be used by human and machine labelers.
2018-2021For a brief period, I consulted with The Innovation Group.
Leading up to Oceans Casino's relaunch, I prepared some research on the market segments, and designed their loyalty program.
2018I conducted surveys and matched to Census data to model demand. I used this in a gravity model to estimate market size of sports betting in states that were considering legalizing.
2018I was a manager in a group that analyzed about $500M of marketing budget; as manager, I drove the direction/workflow. We touched many branches of marketing, including direct mail, host program, events/promotions, loyalty program, and advertising. I worked on a wide range of projects, including: Ad hocs; building reports; A/B split testing of DM campaigns; goal-setting for casino hosts; market segmentation; advertising impacts; and test-analyzing survey results.
Conducted ad hoc and strategic analyses on marketing effectiveness and consumer behavior (e.g., digital direct mail, Asian play trends, cross-property marketing), delivering executive-ready insights and recommendations.
2016-2018Integrated and reconciled player and marketing data across multiple source systems, improving data completeness and reliability for downstream analytics.
2016-2018Advised on statistical methodology and developed software to quickly run AB testing and reporting, resulting in an 80% reduction in process time. Reporting included visualizations for drill down decision-making. Created results repository for long-term trend analysis.
2016-2018Redesigned predictions of hosted players' play to improve transparency and interpretability, achieving more accurate identification of high-value players for ~80% of cases and enabling better rewards allocation.
2016-2018Developed consistent market segmentation methodology across 15 casinos, including spend-per-trip and visit frequency buckets, reconciling differences in underlying metrics to ensure comparability.
2016-2018Built a Tableau dashboard to show event KPIs versus benchmarks. Spearheaded initiative and achieved wide roll-out to about 60 users at 15 casinos with 1000s of uses per month, becoming most-used workbook in the company.
2016-2018Performed sentiment analysis on year-end survey results, and communicated results to executive leadership.
2016-2018Conducted a live game theory experiment to help decide how to bid for gaming licenses.
2016-2018My team built the models for Auto Owners' commercial line products, including TTP, commercial auto, workers comp, and others. My work was divided about equally into three tasks: Data work, modeling, and research. Data work was SQL work to pull data for our models, and the models were large general linear models.
Used an SVM on text to predict fraud from claim notes. This allowed us to automate the work of 15 FTEs.
2014-2016Reversed-engineered pre-packaged GLM software, allowing us to automatically produce modeling packets. This reduced a day-long project to minutes.
2014-2016Researched and advised analysts in the company on dimension reduction in auto and credit datasets. We looked into PCA, partial least squares, and lasso regressions.
2014-2016Built a customer lifetime value model of our commerical policy data.
2014-2016Helped refresh our decade-old commercial auto model, combining two previous models.
2014-2016I presented research on Shapley values and family errorwise rates that influenced our modeling techniques broadly.
2014-2016I thought Intro to Stats at nights one summer while I was an actuary. That semester I redesigned the term project.
I was the company's only underwriting actuary. My main job was to create and maintain software to renew group insurance policies. Additionally I worked on a number of small and ad hoc projects.
Researched to understand how new legislation impacted the way that we priced policies.
2014Wrote a web scraper to get a ICD-9 to ICD-10 crosswalk.
2014Priced our new small ASO product using Monte Carlo simulations.
2014While in grad school, I thought a dozen classes over six semesters. These classes included algebra, math for education majors, and calculus 2 and 3. As teacher, I taught classes; met with students; wrote and graded tests; and reported grades. In my first year, I won a reward from the department for teaching.
GPA: 3.83
Passed qualifying exams on geometry/topology, algebra, and analysis.
2011-2013I won an award for best junior teaching assistant.
2011-2013GPA: 3.83
As an undergraduate, I won first place at UNR in each of: The Putnam exam, the Intermountain Mathematics competition, and the university’s Association for Computing Machinery programming competition. My team of three won the designation of meritorious winner in the international COMAP Mathematics Competition in Modeling.
2007-2011I participated in a number of clubs, as well as founding UNR's math club and go club.
2007-2011