Justin Wei

Built and scaled an AI-powered search engine for developers from pre-product to millions of searches a week. Reached $800K ARR and raised a Series A ($10M total). Set weekly product direction for a team of four engineers based on feedback from usage data, Discord, in-person events, and email outreach. Deployed custom LLM inference on bare-metal GPUs with the TensorRT LLM library, handling 20 inferences per second under 300ms latency. Configured an auto-deploy CI/CD pipeline for a React / Next.js web app using AWS ECS, Cloudmap, Terraform, load balancers.

Kalshi is developing a new asset class, “event contracts,” and a financial exchange for trading on the outcome of events. On track to be federally regulated by the CFTC as the first-ever legal futures exchange in the U.S. for trading on a broad category of events.

Global Quantitative Strategies Alpha R&D Team - Worked closely with execution quants to build a performant analytics platform. Commodities Team - Created and deployed a statistical model on the international flow of cocoa.

8th engineer at Sequoia-backed startup building end-to-end encrypted collaboration tools, e.g. ultra-secure GoogleDocs. Investigating best practices in secure authentication, authorization, session management, and cryptography.

Developed a translation between the SCSI and NVMe storage protocols, allowing future standardization to reduce maintenance costs by 3x. Worked with a team of four to design a system covering hundreds of pages of technical documentation.

Implemented a write-through Redis database cache, fixing scalability and data consistency issues while maintaining sub-100ms API response times. Connected API to a centralized reporting system using Datadog metrics and Splunk log forwarders.

Created test cases in a CI/CD pipeline to provide full automated coverage of 12 core APIs, saving 100+ hours per year on manual testing. Parallelized tests against a local database to reduce test suite running times by 95% (12 min -> 40s).

Trained machine learning models on traffic stop data to predict the outcome of vehicle searches. Improved the accuracy of a SVM classifier by over 30% using the scikit-learn library.

Developed a web-based visual analytics tool for exploring data from academic conferences using d3.js. Published and presented an academic paper at the Visualization and Data Analysis 2018 conference.

S22 Batch