Hello!

Email: ude.yelekreb@leen

About Me

Neel Somani

I'm a quantitative analyst at Citadel in Chicago, IL, working on the Central Research team in Commodities. I graduated from UC Berkeley with a triple major in computer science, mathematics, and business administration in May 2019.

My research projects focused on type systems, differential privacy, and highly-scalable machine learning systems. Fun fact: I'm a licensed Realtor with the brokerage Keller Williams!

Industry

  • Citadel (Quantitative Research Analyst ← Quantitative Developer): As a developer, I built research tools like the model execution framework and the distributed compute system for the Commodities organization.
 Sep. 2020 - Present
  • Airbnb (Software Engineer): I worked on the Growth Foundation team to attribute bookings to ad campaigns and allocate spend to maximize ROI. I also launched new features and campaigns for our Referrals and Affiliates programs.
 Aug. 2019 - Sep. 2020
  • Two Sigma (Software Engineering Intern): I built a Python library to easily estimate the expected return and volatility of various economic and financial factors over time.
 May - July 2019
  • Bain & Company (Associate Consultant Intern): I worked with a large biotechnology company on exposing the cost drivers of clinical trials. My work implemented more accurate budgeting practices in over 500 cost centers globally.
 Jun. - Aug. 2018

Selected Research

  • Cirrus: I worked with Prof. Randy Katz in the UC Berkeley RISELab to build a highly-scalable machine learning framework that runs in a serverless environment using AWS Lambda.
 2019
  • Duet: I worked with Prof. Joseph Near and Prof. Dawn Song to build privacy-preserving services using dedicated hardware (Intel SGX) and machine learning algorithms. Duet is a type system that allows developers to check their code for differential privacy, statically and automatically (paper).
 2019

Selected Projects

  • Literature card game bots: I write technical blog posts in my free time. Here's an example of a post where I train a neural net to play a card game called Literature using a modified implementation of Q-learning. I later built an open-source React app for the game.
 2020
  • Brainspell: This is a statistical platform for the human curation of neuroimaging literature, which I worked on with the Berkeley Institute for Data Science. The project was presented at the 2018 Organization for Human Brain Mapping conference (abstract).
 2018