selected work
projects
a closer look at products, experiments, and tools i have built across ai, data, and product engineering.
project 01
Dyno
Local-first AI Inference SDK with Cloud Fallback
Dyno is an SDK and runtime system that helps developers offload AI inference to users' own devices when possible, while falling back to cloud providers when needed. The system includes a TypeScript SDK, local agent, hosted control plane, and dashboard for configuring execution policies, device readiness, and model routing. Dyno is built around the idea that AI apps should be cheaper, faster, and more privacy-preserving by using local compute whenever it is reliable enough.
project 02
Multi-GPU AI Server
Self-hosted Local LLM Infrastructure
I built a multi-GPU local AI server to run open-source language models on my own hardware. The system uses repurposed GPU mining hardware with vLLM, Ollama, and Open WebUI to serve models through OpenAI-compatible APIs. I configured multi-GPU inference, Docker-based deployments, remote access through Tailscale, and model serving for experiments with local AI agents, private inference, and cost-efficient development workflows.
project 03
HFT-LLVM
Compiler Optimizations for High-Frequency Trading Systems
HFT-LLVM is a compiler optimization project focused on improving the performance of latency-sensitive trading workloads. I built a realistic C++ limit order book benchmark and implemented LLVM passes that optimize hot-path execution by improving cache locality and removing unnecessary atomic overhead in single-threaded production settings. The project explores how domain-specific compiler transformations can make systems code faster without changing application-level logic.
project 04
HappyHr
Dynamic Price Adjustment and Analytics for Small Businesses
HappyHr is a startup I’m building to help small businesses maximize revenue through dynamic pricing and smart analytics. Inspired by a late-night line at Joe’s Pizza in Ann Arbor, the idea was born from a simple question: would paying a little more during peak demand stop customers from buying? This summer, I’m diving deep—conducting customer interviews, researching surge pricing strategies, and building an MVP to bring real-time, data-driven pricing to restaurants and retailers.
project 05
Chopsticks
AI-powered Video Editor
Chopsticks is an AI-powered video editing platform built for creators. Designed to streamline long-form content editing, it features a chat-based interface for intuitive video manipulation and an automated clip generator that identifies and extracts the most engaging moments. By blending deep learning with a seamless user experience, Chopsticks saves time, boosts productivity, and helps creators grow their channels faster.
project 06
HiveHunt
AI-powered Real Estate Assistant
HiveHunt is dedicated to enabling users to confidently represent themselves in real estate transactions, providing the tools and resources necessary to navigate the home buying or selling process independently.
project 07
PupPod
ML-based Web App for PHD students to analyze dog behavior
Led the web development and machine learning teams to implement a dynamic web application for PhD students, facilitating advanced data analysis and improving research efficiency by 40% when using calculated automated insights. Collaborated with co-founder to develop a MySQL data storage system managing information for 5,000 dogs and 130,000 game sessions. Created a custom GUI to streamline data retrieval increasing interpretability of complex datasets.
technologies
project 08
Earning Calls Stock Predictor
a tool to help track expenses and plan budgets
Created an intuitive algorithm, using scikit-learn and open-ai, to analyze investor reports and predictions for all relevant stocks before an earnings call, buy x amount of the stock 1 day before the call, and sell 1 day after. Resulted in a 15-20% ROI in April 2023 - May 2024, on a month to month basis. Scraped Zacks Investment Research to get/store earnings call data in a MySQL schema. Used data to train regression model. Created GUI using tkinter to display useful inferences made and change metrics used to judge data.
project 09
personal website
a minimalist personal website with a purple theme
this is my personal website built with next.js and tailwind css. it features a clean, minimalist design with a purple color scheme. the site includes information about me, my projects, and ways to contact me.