|
|
Engineered 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.
2021-2022Designed 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.
2018-2021