// mathematics · computer science · cornell '28

Justin Roberts

Hi! I'm an ex-SWE intern at Wayfair who's a little too obsessed with LLMs (cough cough — my Chrome extension, this website, my resume, etc.). If I'm not coding I'm probably rock climbing, playing chess, attempting to cook, or taking a nap. Still figuring out summer 2027 plans — feel free to reach out!

jbr272@cornell.edu GitHub ↗ LinkedIn ↗ Ithaca, NY · (631) 759-1717
01 — experience

Software Engineer Intern · Wayfair

Jun 2026 — Aug 2026 · Boston, MA
  • Developed a Gemini-based review moderation system on a BigQuery → Vertex AI pipeline with Airflow DAGs — ~1K reviews/day at 89% precision, 98% recall against human moderator decisions.
  • Built an end-to-end image upload & retrieval pipeline in Java/Spring Boot with GraphQL and GCS signed URLs — 10K+ uploads/day at sub-500 ms p95 latency.
  • Optimized PostgreSQL ingestion for 5M+ reviews/year, cutting redundant UPDATEs by 80% and downstream Airflow events by 60%.
  • Led a Devin AI knowledge-transfer session for 15+ engineers on agentic debugging workflows, cutting team PR rollback rate by 30% within weeks of adoption.

Software Engineer Intern · PlayToon

Sep 2025 — May 2026 · Remote
  • Redesigned a Prisma ORM schema in JavaScript, reducing data redundancy by 30%.
  • Built an AWS S3 upload pipeline across dev/staging/prod comparing presigned-URL vs. server-mediated flows — 40% fewer failed uploads, 50% simpler IAM.
  • Overhauled developer docs with setup guides and architecture diagrams, cutting onboarding time by 30%.

Teaching Assistant · Intro to Machine Learning, Cornell

Aug 2026 — Present · Ithaca, NY
  • Deliver weekly office hours and discussion sections for 300+ students, translating core ML theory (SVMs, gradient descent, neural networks, regularization) into applied intuition.
  • Author and grade problem sets and exams, calibrating rubrics with 5+ course staff.

Independent Researcher · Stony Brook University

May 2025 — Oct 2025 · Stony Brook, NY
  • Developed sublinear graph algorithms using global queries to detect isolated vertices — logarithmic query complexity, 75% fewer queries than prior work.
  • Validated performance in Python across 1,000+ synthetic graph datasets, 25% faster than legacy implementations.
02 — selected projects
JavaScript · Express · Chrome Ext. API · Claude API

Buzzword Decoder

A Chrome extension (Manifest V3) with a Node/Express proxy on Render, integrating Claude Haiku 4.5 to translate corporate jargon in real time — sub-2s end-to-end latency, a whitespace-normalization algorithm mapping collapsed text back to original DOM offsets, and an API key secured behind a token-authenticated proxy.

<2s
end-to-end latency
~35%
phrase-match accuracy ↑
0
CSS conflicts / 100+ sites
Buzzword Decoder tooltip translating 'ontology-powered operating system' into plain English
Python · yfinance · Alpha Vantage · ccxt
code ↗

Trading Signal Engine

A Python engine implementing SMA, Bollinger Bands, and RSI indicators for automated buy/sell recommendations, fed by a market-data pipeline spanning equities and crypto.

+21.50% BTC +14.39% ETH +16.29% BNB
Python · Graph Theory · Research
code ↗

Sublinear Graph Algorithms

Isolated-vertex detection using global queries, achieving logarithmic query complexity — a 75% reduction over prior approaches — validated on 1,000+ synthetic graphs with 25% faster runtime.

Stony Brook University
03 — toolkit
Languages
Java Python C JavaScript SQL GraphQL Assembly Julia HTML/CSS PHP
Frameworks
Spring Boot React Node.js Express FastAPI Django PyTorch
Tools
AWS GCS Docker Kubernetes Git GitHub DataDog Buildkite Chrome Ext. APIs Anthropic API
Libraries
pandas NumPy Matplotlib ccxt yfinance Alpha Vantage