Sean Wade

• Architected and built a Python library for LLM-based quality evaluation framwork, enabling parallel execution of prompt DAGs and automated quality evaluation metrics
• Researched novel techniques in meta prompts and agentic workflows; built testing and validation suite to evaluate AI rater quality at scale and measure outcome variability
• Created metrics to improve Google Discover recommender system, validating retrieval algorithm quality and performance
• Provided technical guidance on experiment design and analysis, identifying meaningful lift across key user sub-cohorts

• Led cross-functional teams in developing, shipping and maintaining core on-device ML algorithms to users:
  • Meal logging tool with food image classification and LLM nutrition feedback
  • Personalized workout recommender system using a multi objective loss function and online learning
  • Unsupervised walk detection algorithm from steps data
  • Contextual notifications engine using routine identification (commute to work, typical workout time, etc)
• Evaluated performance of ML models with focus on safety, robustness and privacy and efficacy communicated tradeoffs to partners
• Architected data schema and privacy preserving data collection framework for launch of multiple health related programs and research studies (Lumihealth, Heartline, Attain)
• Created application for labeling health data to create datasets and provide human feedback for models
• Built data pipelines to process high volumes of medical claims data and health sensor data
• Designed research studies and experiments to study impact of digital interventions and models on hypertension, diabetes, asthma, MCI and atrial fibrillation

Researched synthetic cancer cell image generation using conditional adversarial neural networks and wrote a python library to distribute it. Created novel methods for representing medical claims history with vector embeddings.

• Created tools to improve deploying, evaluating, and retraining ML models on Azure
• Extended Azure SDK to support MLflow

• Developed recommender system for rides/attractions in Disney World Park app
• Created realtime pipelines and dashboards to monitor performance

• Used drone imaging and photogrammetry to construct 3D models of buildings
• Built convolutional neural network to identify and classify roof damage