AdTech AdTalk Ask an Expert (ft. Tylynn Pettrey, Chalice AI)
Jan 30, 2026
Tylynn Pettrey, SVP of Data Science at Chalice AI, leads work on predictive modeling, deep learning, and low-latency ML for advertising. She traces modeling history from linear regression to transformers. Short takes cover sparse ad data, embeddings, model explainability, agentic AI hype, quantum in ad tech, practical hiring skills, and a quirky mass spectrometer pandemic story.
AI Snips
Chapters
Books
Transcript
Episode notes
Transformers: Attention Enables Contextual Memory
- Transformer self-attention remembers token relations and parallelizes processing, enabling modern LLMs.
- Tylynn highlights this breakthrough as key to text understanding at scale.
Imbalanced Data Is Core Ad‑Tech Challenge
- Ad‑tech faces extreme class imbalance: millions of impressions but few conversions.
- Tylynn outlines undersampling, oversampling, and synthetic data (generative AI) as balancing strategies.
Prune And Engineer Features Before Modeling
- Practice feature selection and engineering to avoid multicollinearity and excessive memory or compute costs.
- Combine, transform, and reduce variables so models use only the most predictive, non‑redundant features.
