Building yobi.aiWe are hiring ML Engineers across the skillset spectrum.Reach out if you're interested in at least a few of {representation learning, recommender systems, data engineering, build systems, building 0 -> 1 products and systems, adtech, graph neural networks, causal inference, ...}

Tech Lead of the User Modeling Research team @ https://cortex.twitter.com/en.htmlThrough ML and product innovation, our team was able to drive millions of monetizable daily-active-users, increase content product, launch novel "producer" side experiments, design brand new products to facilitate "People You May Know" recommendations and so on.Fun fact: Promoted to Senior Staff the week that I quit.My work starts from the end-user/consumer problem and typically involves using graph based methods on our large unstructured social datasets.Sample projects:* Friend Finding (People You May Know) ecosystem work. Includes models that power the recommendations as well as new products to help better leverage those models.* Large scale user similarity computation. Powers many recommendation sources.* Large scale semi-supervised node classifcation framework.

* Unsupervised learning on multivariate time series (Toeplitz Inverse Covariance Clustering, Hidden Markov Models, etc.)* Built streaming graph DSL (read: internal 'pandas' or 'dplyr') to build up complex real-world scenarios from sensor data* Using all of the above to move the organization to scenario based reporting to establish the safety case. Log level or trip level metrics fail to adequately describe the true operational domain of driving a vehicle.

Large scale distributed data collection (10k QPS).Some high level projects:* Time Series Interpolation microservice* Generative deep learning model for novelty detection* Rewiring an 'eventually consistent' cronjob (1 TB per day) ETL to a proper DAG

* CI/CD real-time Machine Learning (Python and R models) deployment tool built on Docker. We achieved 13ms latency when enterprise tools were in the 80-100ms range at the time.* Marketing Attribution - Structural time series models to evaluate non randomized ad campaigns.* Ad CTR model - The usual stuff.* Bayesian AB Testing - Moved org over to Bayesian formulae/methods including hierarchical partial pooling and posterior loss annealing to handle multiple testing and peeking respectively.

Part time instructor for GA Data Science course.

Algorithmic trading of oil futures.