Gil Pasternak

Member of Technical Staff at Fastino • NLP Researcher • Interpretability Enthusiast

About Me

Gil Pasternak

Hi, I'm Gil

Currently a Member of Technical Staff at Fastino, where I do foundational Language Model research. My interests sit at the intersection of natural language modeling, interpretability, and (some) Machine Learning theory.

Previously, I was a Researcher at Loris (acquired by Contentsquare), one of the earliest core engineers at Great Expectations, and did AI research at NVIDIA. I earned my Bachelor's in Mathematics and Computer Science and Master's in Computer Science from UC San Diego, where I was fortunate to be advised by Misha Belkin and Ramamohan Paturi.

Outside of work, you can usually find me reading mechanistic interpretability papers, lifting heavy things, or getting into lengthy conversations with Uber drivers. I love chatting about interpretability, language models, or the weird quirks of language-based systems — if this is something that also interests you, reach out!

Experience

Fun / Projects

AgingGAN

A GAN trained for a Deep Learning course that takes faces and ages them. Includes both a custom U-Net Variation (designed from first principles) and a modification of CycleGAN. Also includes some interpretability work, including Guided Backpropagation and Feature Visualization.

View Project

Likelihood

An ML-Centric Data Quality Engine trained to identify anomalies in data, irrespective of domain. Includes a multitude of methods, including KDE Estimation, Time-Series based prediction, and PCA.

View Project

SongRecommender

A recommender system (collaborative filtering + content based algorithm based) that beats some existing baselines for song recommendations, equipped with a smooth UX design.

View Project

Contact