Noa Bendit-Shtull

Noa Bendit-Shtull | ML @ Etsy

I am an experienced data scientist with a passion for diversity in tech, efficient code, and effective communication. I care about how work gets done, and lead by example.

As an Applied Scientist at Etsy, I apply machine learning methods to build Etsy’s Knowledge Base, a repository of structured information about Etsy’s marketplace. Previously, at Oscar Health, I built machine learning models to systematically detect and diagnose excess medical spending.

Aside from machine learning, I am passionate about teaching and mentoring; at Stanford and U. Penn, I was an economics teaching assistant both online and in-person. At work, I’ve developed and led training sessions on technical and non-technical topics (e.g. git, R, giving feedback), and have served as a formal and informal mentor.

When I’m not at work, you can find me baking pies, glass blowing, or swing dancing.

Programming

I write elegant, efficient, and readable code in Python, R, SQL, C++, and more. I work collaboratively to build testable, well-documented code bases.

Data Wrangling

I draw on my four years of experience in consulting to untangle messy data and suss out anomalies.

Statistics

I leverage my masters education in Statistics to identify the right techniques for each setting.

Data Engineering

I construct robust data pipelines to parallelize the processing of large datasets.

Communication

I communicate technical content to non-technical audiences to make data science clear and approachable.

Teaching

I have 8+ years of experience teaching in both academic and corporate settings, and love every minute of it.

Argument Visualization

Causal Inference

My team conducted a reanalysis of the data from a published study using causal inference techniques to account for the lack of randomization in the experimental design.

Second Order LSTMs

Natural Language Processing

We test the hypothesis that second-order LSTM architectures have the improved ability (relative to a simple LSTM) to learn long distance dependencies within an input.

Dance Classification

Video Action Recognition

We assess the ability of sequential neural networks to capture dynamic movement and predict dance styles from video.