DylanSatow

Rising Senior at Columbia studying Math and CS

Dylan Satow playing trumpet on stage with warm lighting

Dylan Satow playing trumpet on stage with purple lighting

Dylan Satow playing harmonica under a red tarp outdoors

About Me

I'm Dylan, a rising senior at Columbia University. I'm super into broadening my horizons in both math and computer science, with my current interests being Mechanistic Interpretability and Representation Learning. I also love more general CS/Software Engineering, and have recently gotten much more into high performance computing (HPC) and web development. I am also an avid jazz trumpet player/multi-instrumentalist, with experience playing at many notable venues in NYC

Experiences

Software Development Engineering Intern

Amazon Web Services (AWS) - Santa Monica, California

Summer 2025

SDE intern on the Amazon MSK team (Managed Streaming for Apache Kafka).

Pe'er Lab Research Assistant

Columbia University - New York, New York

January 2025 - May 2025

Non-Euclidean Representation Learning and Genomics

DAP Lab Research Assistant

Columbia University - New York, New York

November 2024 - May 2025

Bi-directional Causal Discovery for Relational Databases

ReAIM Lab Research Assistant

New York Presbyterian Hospital/Columbia University - New York, New York

May 2024 - October 2024

Transformer-based Zero-shot Bayesian Inference (Prior Fitted Networks) for EHR data

Nakul Verma's Research Assistant

Columbia University - New York, New York

January 2024 - December 2024

Unsupervised Learning Theory & Creating Tools for Neuroscientists

Data Science Intern

Bonafide Health - Harrison, New York

June 2023 – August 2023

ML for predicting Customer LTV and CACs

Projects

AI Lab Management Software (withixion.com)

AI-powered lab management software currently being used/adopted by 3 Columbia University labs.

NextJS

Django

LangGraph

AWS

View Project

CUDA Neural Network from Scratch

Cuda Neural Network from complete scratch (no DL/Math libraries)

CUDA

C++

Neural Networks

GPU Computing

View Project

Manify Coauthor

Non-Euclidean Representation Learning Library/Paper. 50+ stars on Github

Python

PyTorch

Differential Geometry

Representation Learning

View Project

Synthetic Time-Series Generation for Imbalanced Data

Implemented conditional TimeGAN to address severe class imbalance (10.94:1 ratio) in occupancy detection using PIR sensor data. The solution generates high-fidelity synthetic time-series data that preserves temporal dependencies while oversampling minority classes, improving classification performance on real-world imbalanced datasets.

DL/GenAI

Time-Series Analysis

Class Imbalance

MLOps

Research Implementation

Model Evaluation

View Project

Multi-Answer Speculative Decoding using Ngrams (Streaminator)

Final Project for COMS E6998 High Performance Machine Learning. Inference pipeline that uses per-prompt n-gram speculators and continuous batching to optimize multi-answer generation with adaptive token prediction.

CUDA

Python

Pytorch

HPC

NLP

View Project

LexiLocal: AI-Powered Legal Document Summarization and Q&A

LexiLocal is an AI-powered legal document analysis application that provides intelligent summarization and Q&A capabilities using local LLMs for maximum privacy.The system processes legal documents through a RAG(Retrieval- Augmented Generation) architecture with semantic search and optimized quantized models for efficient inference.

Python

LangChain

Streamlit

FAISS

Sentence Transformers

Ollama

View Project

Selected Coursework

High Performance Machine Learning

Graduate

Unsupervised Learning

Graduate

Machine Learning

Graduate

Natural Language Processing

Graduate

Modern Analysis 1

Graduate

Modern Algebra 1

Graduate

Music

Experience

Long time jazz trumpet player, and (occasional) player for Noshows (go stream it on all platforms). I dabble in EWI (Electric Wind Instrument), ocarina, guitar, bass, piano, and synths. I'll add some demos here if I'm not lazy...