John Schultz

• GPA: 3.7. Coursework: CS 61B (Data Structures), CS 162 (Operating Systems), CS 170 (Algorithms), CS 61C (Computer Architecture), CS 189 (Machine Learning), EECS 151 (Digital Design and Integrated Circuits).
• News and Data Reporter, Daily Californian. Member, Political Computer Science. Individual Contributor, Berkeley Tech & Justice Lab

• Built a Kubernetes/AWS resource layer for DeepTS, the production deep learning library that powers training and batch inference for billions of Amazon forecasting decisions, improving reliability and resource utilization of distributed compute.
• Ideated and shipped an internal monitoring service for distributed training runs, giving scientists a single place to diagnose time-series training jobs over many services and runtimes.
• Built telemetry quantifying cross-org use of DeepTS, used by leadership to inform business decisions.

• Designed and trained an ML model (text embeddings, XGBoost, Census data) to predict customer adoption of Add To Delivery, an Amazon delivery feature.
• Engineered a Spark-based data ingestion and feature engineering pipeline to create replicable data for model training.
• Shipped a Java/React frontend to surface model predictions and counterfactuals for feature adoption, used to inform rollout strategy.

• Built automation features with Java and Python to validate data submission from 71 federal services across 38 agencies.
• Prototyped NLP analysis of customer reviews for federal agencies, presenting topic-based sentiment analysis results to a large partner agency.

• Designed a featurization pipeline to predict eligible cash aid recipients in Togo from call detail records, raising model accuracy from the lab’s prior baseline.

• Worked on GradeView, an application to represent CS10’s unique mastery-based learning system to students.

Created a web application to map the recovery of AC Transit, the East Bay’s transit agency. The data is simplified from a 50GB response to a public records request I made. Includes origin-destination estimation and data imputation to cover missing data. Made with Next.js and Python. Used Google Cloud instances, storage buckets, and Cloud Run to process and serve data.

The Berkeley Tech and Justice Lab creates tools to help law and justice advocates perform their jobs. Many stakeholders file public records requests, and need a way to organize requests and draft verbiage for letters sent to public agencies. PRA Machine is a web application that uses an LLM to draft request letters and implements an agentic inbox to track requests.

Built a React and Flask mobile app that finds empty rooms to study in at the UC Berkeley campus. Scrapes Berkeleytime, a third-party application with access to UC Berkeley class catalogs. Includes a SQLite database to track rooms taken not in the class catalog for all users.