Projects
A few things I’m proud of.
Across infrastructure, AI systems, and products. Some shipped to enterprises, some to 50,000 users, some only to a research server. All of them taught me something.
Side Project · 2026
Paul Code
A coding agent I built to better understand how AI agents work under the hood. Exposes a small, sharp toolkit — read, write, edit, grep, glob, bash — and runs the typical engineering loop: read the relevant files, propose a focused change, verify it.
Sunrise Investment Club · 2026
Sunrise Investment Club Course
Full-stack course platform for the Sunrise Investment Club. Authenticated lessons, a rich-text editor for course content, internationalization, and a Postgres-backed schema — designed and shipped end-to-end.
UPenn · Fall 2024
Orca LLM Serving System
Re-implementation of the Orca distributed serving paper. Achieved 36.9× higher throughput than FasterTransformer using iterative scheduling and selective batching, and rewrote the OPT transformer to parallelise non-attention operations.
Co-Founder & Interim CTO · 2020 — 2022
Fanita Health
Telehealth platform for stigmatised healthcare in Indonesia. Built the prototype, scaled to 50,000 users, hired a team of ten, and raised a six-figure pre-seed round. The product taught me that systems only matter if they reach the people they're for.
Pattern Inc. · 2024
LLM Routing & Cost Reduction
Fine-tuned BERT to classify and route prompts to the cheapest viable model — cut LLM spend 85%. Separately, replaced parts of the image-processing pipeline with AWS Rekognition + Airflow, dropping OpenAI costs another 45%.
UPenn Research · 2023
Voice-Enabled AI Tutor
Full-stack LLM platform teaching English to students in India. Cut chatbot latency 80% via response streaming and prompt engineering, and designed the system architecture end-to-end.