Built a cluster scheduler sustaining 500+ concurrent tasks across 10 nodes via leader-elected coordination, heartbeat-based failure detection with sub-3s failover, bin-packing allocation, and automatic task reassignment.
Implemented a multi-tenant, developer-facing event ingestion and processing platform sustaining 10,000 RPS, with API-key auth, rate-limited and idempotent intake, asynchronous pipelines, dead-letter isolation, and observability.
Engineered a Go and gRPC fanout-on-write feed-ranking service generating personalized timelines for 50,000+ simulated users, then added a Redis caching layer that cut database read load 80% while sustaining 8,000+ requests/sec.
Architected a vectorized, single-node OLAP engine in Go querying Parquet via Apache Arrow, with predicate pushdown, column pruning, and row-group skipping, sustaining 100M+ rows/query at 10x a row-at-a-time baseline.
Built a production-style LLM inference server in Python and FastAPI with continuous batching, KV-cache reuse, and request queuing, sustaining 200+ concurrent requests while cutting p99 latency 45% and lifting throughput 3x.
Developed a Dockerized robotics control system that gives an AI agent full control of a robotic arm: the agent generates, validates, and executes low-level Python motion policies in a sandboxed ROS layer, using only minimal motion APIs (no skill library or demos) so it can handle unseen tasks in one continuous loop.
Rebuilt Hornslink from scratch using React Native and Flask in direct collaboration with UT Austin, modernizing the application used by 50,000+ students to discover campus opportunities.
This site! Next.js and Tailwind CSS, featuring an AI clone of me that you can chat with to learn more about my background and work. You can also ask "me" to schedule a time for a video chat, and "I" will handle the rest!
Automated lead generation with an AI agent that uses LLM tool-calling to identify Instagram client profiles, extract emails, and run targeted outreach campaigns, achieving 2x higher conversion rates compared to industry averages.
Revamped the org's website in Next.js. Replaced a fragmented Google Suite workflow with an internal portal for applications, reviews, attendance, and numerous org-wide logistics.
Created a Chrome extension that eliminates manual scheduling by converting Google Calendar availability into clean, ready-to-send email text with configurable date ranges, meeting durations, and timezone conversion.
Launched an NBA analytics platform tracking 700+ players with 16,000+ stats, featuring advanced filters and ML-powered game predictions.
Partnered with the Texas Parks and Wildlife Department to create a full-stack web-app supporting ecological restoration efforts in Texas, powered by simulations and machine learning.
Designed a mobile app that helps tourists and new residents find and navigate safer walking routes via custom real-time safety alerts.
Programmed an autonomous self-driving car in the Udacity simulator that uses computer vision–based lane detection and traffic sign recognition to enable real-time vehicle navigation. Still kind of gimmicky.
Put together a simple official website for the UT Racquetball team because my friend asked me to (it is fully vibe-coded in a single prompt, lol).