A Bloomberg-style trading simulator where you manage a $10k virtual portfolio against live-streaming prices, with an LLM copilot that analyzes your holdings and executes trades from plain-English requests.

It lets people practice trading and portfolio management with zero risk — no real money or brokerage account — while demonstrating how an LLM can act as an agentic copilot that turns instructions like “trim my tech exposure” into validated, executed trades.

A single FastAPI service serves both the JSON/SSE API and a Next.js front end. A Geometric Brownian Motion simulator (or the Polygon API) streams prices to the browser over Server-Sent Events, with state in SQLite. The AI layer uses LiteLLM with Pydantic-typed structured outputs, and any trade it proposes runs through the same server-side validation as a manual one — the model proposes, the code enforces. Containerized with Docker.

An AI-assisted scheduling app for restaurants: define roles, employees, skills, and availability, then generate, edit, and publish staff schedules in minutes — with time-off requests, shift swaps, templates, and clock-in/out attendance.

Building restaurant rosters by hand is slow and error-prone — juggling who's qualified, who's available, approved time off, overtime, and coverage. Mise drafts a compliant schedule from those constraints in seconds and lets managers adjust and publish to staff.

A monorepo: a FastAPI backend (SQLAlchemy + PostgreSQL, JWT auth) and a Next.js 16 / React 19 / TypeScript front end. The scheduler assembles a structured context of shifts, roles, and availability and asks an LLM (DeepSeek) for assignments — then deterministically validates every one server-side (role, availability, time-off, no double-booking) before saving. The AI only drafts; the validator decides what actually ships.

A private household organizer that unifies a family's chores and tasks, a shared calendar with two-way Google Calendar sync, an AI “Chef” that plans the week's meals from your recipe catalog, and an auto-generated grocery list.

Families juggle chores, calendars, meals, and shopping across disconnected apps and separate Google calendars. Family-Hub centralizes it for one household, keeps both parents' calendars in sync with a single shared view, and offloads weekly meal planning to an AI agent that respects the recipes you actually cook and your busy nights.

A single Next.js 16 app (App Router, React 19, Server Actions) with no separate API — mutations flow through a service layer the AI tools reuse, so business logic lives once. Data is in Neon Postgres via Drizzle ORM; auth is Auth.js (NextAuth v5) with Google, requesting Calendar scope so sign-in also authorizes sync (refresh tokens encrypted at rest). A Vercel Cron job runs incremental two-way calendar sync every 15 minutes, and the meal planner is a tool-calling agent on the Vercel AI SDK (DeepSeek).

A jungle-themed chore-and-reward app for families. Kids check off daily chores to earn stars and trade them for parent-defined rewards, with an optional “Jungle Math” mini-game for bonus stars. One Expo codebase ships to Android and a hosted web PWA.

It gives families a simple, gamified way to track chores and motivate young kids with a visible star economy, synced across phones via one shared family login. The hosted PWA means iPhone users can add it to their home screen without the App Store, and a parent PIN keeps kids from approving their own rewards.

Expo SDK 54 / React Native 0.81 with expo-router, shipping native Android and a static web build via react-native-web. State is React Context; Firebase provides email/password auth and Firestore, with data namespaced per family (families/{uid}/…) and enforced by Firestore security rules. Star balances update in real time via Firestore snapshot listeners, and persistence is platform-aware (IndexedDB on web, AsyncStorage on native).

Built end to end as part of my self-directed software study: a server-rendered portfolio with a retro arcade intro and a “Digital Twin” AI chat that answers questions about my background in my own voice.

A résumé is static. This turns it into an interactive experience that maps my operations record to the Scrum Master role and demonstrates hands-on technical ability — scoped vision → epic → story like a real delivery backlog.

Next.js 16 (App Router, React 19, TypeScript): static content is server-rendered, with the arcade gate, AI chat, and contact form as isolated client islands. An /api/chat route proxies to DeepSeek with input validation and an origin allow-list, and Web3Forms handles the contact form. Deployed on Vercel.