A hackathon project: a generative-music web app that maps camera-captured objects to instruments with Gemini Flash, composes full pieces with Lyria Pro, and generates music videos with Veo 3.0 — with a live FFT stem-mixer.

• Gemini
• Lyria
• Veo 3.0
• Python
• Web Audio

Ablated four trajectory representations (Cartesian waypoints, Frenet, velocity, acceleration) inside a vendored NeurIPS-2025 flow-matching planner (Tan et al.), training each variant on 1.5k-scenario subsets drawn from an 80k-sample nuPlan-mini cache. On open-loop ADE (average displacement error), Cartesian waypoints performed best — roughly 5x lower error than the velocity- and acceleration-based representations (4.19 m vs ~22 / 20 m).

• PyTorch
• Flow Matching
• Autonomous Driving
• nuPlan
• Research

A real-time prediction-market app (Next.js 16 / React 19 / TypeScript / Firestore) built over a 6-week Agile project, with transaction-safe coin economics, an LLM (OpenAI gpt-4o-mini) that structures free-form bets into verifiable yes/no markets, and a market-resolution state machine.

• Next.js
• React
• TypeScript
• Firestore
• OpenAI

An academic deep-learning project spanning generative modeling and reinforcement learning: a conditional DDPM with a flow-matching head, GAN/VAE families, a from-scratch MiniGPT transformer, and DQN/DDPG agents.

• PyTorch
• Diffusion
• Transformers
• Deep RL
• GANs

An offline-first, event-sourced knowledge system: it captures text (a global hotkey on Windows), organizes it with a local LLM (Ollama) into a semantically linked Obsidian Markdown vault, and auto-generates dependency-ordered plans via a Kahn topological sort. The platform-agnostic core runs fully offline.

• Ollama
• RAG
• MCP
• Python
• Event Sourcing

A meta-harness that wraps an AI coding agent and enforces eight deterministic, fail-closed engineering gates (build, hygiene, layout, format, lint, typecheck, tests, security) via Claude Code hooks. v0 governs Python projects, with a per-language/per-harness seam designed to extend to other stacks and agents.

• Claude Code
• CI Gates
• ADRs
• Python
• Agentic

A self-driving Raspberry Pi 5 vehicle with ultrasonic sensors and a camera, using Rapidly-exploring Random Trees (RRT) and Visual SLAM for goal-directed navigation on flat terrain.

• Raspberry Pi
• RRT
• Visual SLAM
• Robotics
• Computer Vision

A set of embedded builds: an Arduino + Python IR remote control, a sonar object-detection system, and an FPGA-based “Bop It” game written in Verilog.

• Arduino
• Verilog
• FPGA
• Embedded
• C++

A Retrieval-Augmented Generation system over both textual and image data, using LLMs and vector embeddings to extract insights and generate detailed answers.

• RAG
• LLMs
• Vector Embeddings
• Multimodal
• Python

A physics-accurate drone simulation for autonomous flight, control, and environment interaction, trained with supervised and reinforcement learning in Unity ML-Agents.

• Unity
• ML-Agents
• Reinforcement Learning
• Simulation
• C#