Terry Feng

Scientist · ML Engineer · Researcher

About

Terry Feng

Hey there! I am Terry Feng — scientist and engineer. My primary interests revolve around researching and developing utility maximizing strategic agents, and designing markets and mechanisms that enable them - whether the interactions are between humans, AI, or a combination of the two.

My expertise and work sits at the intersection of AI, game theory, mechanism design and statistical estimation. My depth stems from a decade in advertising and real estate marketplaces, where my work spans building deep learning prediction systems for calibrated ads bidding, designing bidding agents, optimizing auction marketplaces, and the statistical inference engines & experimentation designs to understand them. I see these same concepts playing a pivotal role for the economics and systems of future AI.

My work

  1. Sep 2026–SPAR AIResearch Fellow

    Studying long-horizon dynamics, influence, and strategic behavior in human–AI and multi-agent systems from the lens of mechanistic interpretability and dynamical systems. Designing protocols and interventions that steer those systems toward safer, more aligned behavior.

  2. 2021-2026RedditSr. ML Scientist & ML Engineer - Ranking, Bidding and Exp.

    Drove the optimization of the Conversions suite of models and auction bidding strategies - handling sparse, noisy and delayed feedback data and uncalibrated models. Defined success metrics with XFN and built the experimentation and estimation frameworks for the Ads org.

  3. 2018-2021NextRoll / AdRollSr. Data Scientist & ML Engineer - Pricing and Targeting

    Led pricing and targeting models across the platform — pCVR and pCTR, graph embeddings for contextual targeting and fraudulent supply detection, churn, LTV, and ICP — and spearheaded AI ad creatives on GPT-{2, Neo} over a year before ChatGPT was born.

  4. 2016-2018Anywhere Real EstateML Engineer, RecSys

    Built the recommender systems matching homes, buyers, and agents, and led the migration off Oracle RDBMS onto Spark and Airflow that unlocked analytics and data products across the company.