Portfolio

AI, data, finance, and developer-tooling projects from a CMU computer science student.

Building Products That Turn Messy Problems Into Useful Systems

I like projects that connect product taste with technical depth: structured memory, autonomous maintenance, marketing intelligence, and quant finance pipelines.

My recent work is centered on AI-assisted software, data systems, and decision tools. Each project below reflects a different part of that range: full-stack product design, backend orchestration, applied AI workflows, and statistically disciplined modeling.

Featured Projects

Current public work from GitHub and LinkedIn.

  • Orbit

    AI relational memory app for maintaining a network.

    Orbit converts freeform relationship notes into structured, searchable memory. The product supports capture-before-auth onboarding, Google sign-in, Supabase persistence, semantic search, follow-up lanes, and OpenAI-backed extraction with local development fallbacks.

    • Next.js
    • React
    • Supabase
    • OpenAI
    • pgvector

  • TraceBack

    Autonomous maintenance engine for production errors.

    TraceBack ingests incidents from Sentry and GitHub, queues maintenance jobs with BullMQ, coordinates a Maestro/Surgeon/Verifier agent loop, and produces evidence artifacts for human review before fixes become pull requests.

    • TypeScript
    • Fastify
    • BullMQ
    • Redis
    • Cursor SDK

  • Regime-Switching Portfolio Optimizer

    Quant finance pipeline for market-regime-aware allocation.

    This project uses Hidden Markov Models to detect calm versus crisis regimes, then blends Modern Portfolio Theory allocations through walk-forward backtesting with lagged weights, Ledoit-Wolf covariance shrinkage, and performance visualizations.

    • Python
    • HMM
    • MPT
    • Backtesting
    • Finance

  • BrandAware

    Agentic marketing platform scaffold.

    BrandAware is a Next.js and Supabase product foundation for planning campaigns, generating content sprints, managing post strategy, and using brand DNA as context for AI-assisted marketing workflows.

    • Next.js
    • TypeScript
    • Tailwind
    • Supabase
    • AI Workflows

Earlier Builds

Before the current AI/data focus, I used games and small utilities to build fluency with algorithms, interfaces, and shipping finished work.

  • Dungeonite - a randomly generated maze game using depth-first search to create solvable maps.
  • Tetrix - a game project focused on state, controls, and UI polish.
  • GitHelp - an early developer-support concept that helped point me toward tooling products.