Founder & engineer

Building in AI.

Currently in stealth.

I’m Nick. I’ve spent the last decade building machine learning systems, founding companies, and taking ideas from research to production. Previously, I co-founded Booth AI (YC) and taught deep learning at MIT. In 2016, my research trained neural networks to generate code from natural language.

LinkedIn GitHub X
Nick Locascio

Background

A decade in AI.

  1. Founder · Now

    Stealth AI startup

    Building a new AI company. More to share when it’s ready.

  2. Member of Technical Staff

    DoorDash

    Led engineering teams building DoorDash Assistant, helping consumers find restaurants and build grocery carts through conversation. Led development of AI agent harnesses, simulations, LLM orchestration and optimization, and production evaluation systems.

  3. Co-founder

    Booth AI

    Built a Y Combinator-backed company creating AI-generated ecommerce product imagery.

  4. Director of Engineering

    Standard Cognition

    Joined as employee #7 and helped scale the company to unicorn status. Led ML teams and shipped 10+ models for autonomous retail.

  5. Research & teaching

    MIT

    Graduate researcher. Founded and taught 6.S191: Intro to Deep Learning, MIT’s first deep learning course.

  6. Author · O’Reilly

    Fundamentals of Deep Learning

    Wrote about the foundations of building and training neural networks.

Earlier work

Deep learning consulting at Symantec, recommendation systems at Pinterest, and iOS engineering at Fitbit.

Research

Selected papers.

  1. 2017 · Radiology · Co-author

    High-Risk Breast Lesions: A Machine Learning Model to Predict Pathologic Upgrade and Reduce Unnecessary Surgical Excision

    A model combining clinical information and pathology-report text to identify high-risk breast lesions that could be monitored rather than surgically removed. Published online in 2017; print issue in 2018.

  2. 2016 · EMNLP · First author

    Neural Generation of Regular Expressions from Natural Language with Minimal Domain Knowledge

    Early work on neural code generation: trained a sequence-to-sequence model to turn natural-language descriptions into regular expressions, learning directly from examples without hand-crafted rules for the task.

    Code & dataset

On the side

Notes & experiments.

Writing

Essays on health, science, and what I’m learning.

The data lab

My personal health data and experiments, kept in public.