Building an AI-first app portfolio pipeline:• ML-powered, end-to-end ROI optimizations from market intelligence and app discovery and analysis to acquisition, monetization, and personalization• Building and deploying models on Google Cloud (GCP) • Acquiring and combining diverse signals across large datasets for data-driven portfolio building and market intelligence • Big data machine learning pipeline architecture design with python, Vertex AI Pipelines, Kubeflow Pipelines, LLM RAG processes, scikit-learn, BigQuery• Building teams and partnerships; defining processes

Lead the data science team, architecture and implementation of the company's most-ambitious predictive analytics project.• Modeled risk to quantify and explain expected future dollar value outcomes and their uncertainty; this was accomplished with a multi-level model using biased, noisy, sparse heterogeneous data contributed from many sources (numeric, text, categorical, and date types) and customers• Presented new machine learning product opportunities to our CEO and customers• Helped the team implement techniques to apply active learning, semi-supervised learning, iterative imputation to improve the impact of labels used in our classifiers• Primary tools included: scikit-learn, Python, Pandas; BigQuery (SQL) on GCP; Vertex AI

Collaborating with a team of data scientists and engineers to build models, deliver insights, and automate using Python, scikit-learn and BigQuery SQL (Google Cloud).

Applied machine learning (scikit-learn) to model Google Ads portfolio optimization opportunities, resulting in enhanced customer outcomes and incremental revenue opportunity. Envisioned and implemented a novel optimization solution that was a precursor to the release of Google's "Optimization Score": automotive ad agencies see where they rank across many dimensions and visualize headroom across best practices. This helped customers estimate the counterfactual, "how would my store visit metrics have looked, had I optimized any one of Google's best practices?" For instance, a model estimated the impact of improved site speed on store visit conversions while controlling for other factors like ad spend across keywords.• regression modeling (partial dependence and individual conditional expectation) applied in a high-dimensional space to estimate the impact of complex account changes• Produced actionable intelligence from very large datasets, quantifying customer account optimization opportunities across a billion dollar portfolio• ML product pipeline ideation and architecture w/ advanced feature engineering• Pitched AI solutions to C-suite execs alongside the sales team

• increased Google's publisher support capacity through ROI modeling• performed metrics discrepancy troubleshooting for ad statistics• helped publishers with their ad monetizationTo address capacity planning, I proposed and implemented a novel approach - an ROI-forecasting model to predict different outcomes related to the clients we support. This time-series analysis uses ML techniques to predict ROI, given our objectives and finite support budget. Different modeling approaches are combined to reduce uncertainty across dimensions of interest.Experimental projects include project similarity matching and case-agent assignment prediction for customer support automation.

Data feed automation with Python, PHP, MySQL: data analysis was performed in SQL to identify potential new features and detect anomalies. Feed rules were inferred from patterns and history to develop reliable automation.

BienTek was an early adopter of machine learning applied to advertising by matching supply and demand with data-driven advertising. As the founder, I grew this startup to over $100k annual revenue before selling it. Its business model was made possible by developing a custom CRM and applying ML (predictive modeling) to ad bid optimization, where a database tracked the relationships between job opportunities, contractor availability, and contractor skills. With this, I developed a model to estimate the expected value of advertising clicks, given the history of jobs, contractors, contractor skills, keywords, and their relationships.See http://www.brianbien.com/resume/#bientek