Lucas Chen

A harness-agnostic platform for handing software work to coding agents in Cursor, Claude Code or any other harness. The goal, from my spec: maximize architecture-approved, verified engineering output per dollar and per minute of human attention.

• Wrote the v3 spec: missions, workflow templates, task and role contracts, review gates, model routing, memory and fleet security.
• Built the “agent” control-plane CLI in TypeScript: it scaffolds a target repo, validates artifacts against JSON Schemas and runs the claim → start → submit → gate loop.
• Workers never certify their own work: a task passes only with an evidence bundle, independent verification, and spec, quality and architecture review.
• Routes each task by risk to a capability tier (frontier, strong, mid, cheap), mapped to models by policy instead of hard-coded vendors.

My parents had to set a 7:30 a.m. alarm every week to get courts at our local club, so I automated it. Members set weekly schedules in a web app, and a bot books the courts each morning and texts the results.

• The bot logs in at 7:25, builds every reservation, then fires at exactly 7:30 a.m. ET for courts 7 days out, falls back to backup courts and texts results via Twilio.
• TypeScript monorepo (Turborepo): Express + Prisma/PostgreSQL API with Auth0 roles and machine-to-machine tokens for the bot, and a React, MUI and TanStack Query web app.
• Runs in Docker on a Raspberry Pi: CI builds multi-arch images to GHCR, deploys go staging → smoke test → promote with one-command rollback, and Vitest and Playwright suites cover the code.
• Began in 2023 as “Alfred”: a Playwright bot on a cron-scheduled GitLab pipeline, with a React and Express app, Firebase and multi-user Auth0 login, used by 2 clients. In Aug 2026 the bot moved to the booking platform’s new GraphQL API.

Co-op with Douglas Chen.

A mobile-first meal planner: choose meals and servings, turn them into a grocery list, and track calories and protein against personal targets. Our team of five built it as a front-end prototype on mock meal data, so the “AI-powered meal prep” step is a UI concept.

• Set up the React + Vite app with Auth0 login, the start and home screens and bottom-tab navigation, and deployed it on Vercel.
• Built the meal-planning flow: favourites, filtering out meals that clash with saved allergies and diets, a meal-details view, and meal and serving selection.
• Teammates built the grocery list and meal data (Mohammad Mohsin Khan), the calorie and protein tracker (Wajdan Faheem), and profile and target settings (Jasmine Sun-Hu).

Co-op with Mohammad Mohsin Khan, Wajdan Faheem, Jasmine Sun-Hu, Hammad.

A mobile-first prototype for finding study spots on McMaster’s campus: search and filter spots, save favourites, mark places visited, rate them, and see them on a map around you.

• Leaflet map centred on the user’s location (falling back to Thode Library), with a marker and rating for each spot.
• Text search, attribute filters, and spot pages where you can favourite a spot, mark it visited and leave a rating.
• Runs entirely in the browser: accounts, favourites and visits persist in localStorage. Deployed on Vercel.

A password manager with Auth0 sign-in, an Express API, and passwords encrypted before they are stored in Firebase. I built it in one day with Willie Pai and Mohammad Mohsin Khan for the MEC 2024 programming challenge.

• Added Auth0 login and logout, plus a client service layer that attaches the user’s access token to every API call.
• Added JWT verification to the Express API and set up its Vercel deployment.
• Connected the dashboard to the API for listing, adding and deleting passwords, with password-strength rules on new entries.
• Merged 13 of the team’s pull requests during the event.

Co-op with Willie Pai, Mohammad Mohsin Khan.

An Express + MongoDB API that serves my resume as JSON, one section at a time or all at once. Inspired by Eric Lake’s API resume.

• Endpoints for experience, projects, education and contact, a combined /api/resume, and a menu at /api.
• Only answers Postman (other clients get a 403), rate-limits each IP per day (429), and meets any non-GET request with a joke 500.
• Express and Mongoose on MongoDB, deployed to Vercel with Vercel Analytics.

CarpoolX matches riders heading to the same destination so they can share a taxi and split the fare, with an upfront fare estimate and safety checks for riding with strangers. Four of us built it as a React Native Android app on a Flask and Firebase backend.

• Led the repo: scaffold, navigation and a feature-branch workflow. Merged all 38 pull requests that landed and wrote 119 of 156 commits.
• Wrote most of the MVC Flask backend: multi-user Firebase Auth sign-in, tokens kept in the Android Keystore, token-checked endpoints, and the controller and model layers.
• Built login and registration, the profile, taxi verification by door code, taxi details and passenger reports.
• Teammates built ride requests and matching with Google Maps routes and fare estimates (Henushan Bala), end-of-ride ratings (Rafeed Iqbal) and private group rides (Allison Cook).

Co-op with Henushan Bala, Rafeed Iqbal, Allison Cook.

A small ambient-sound web app inspired by ridethesubway: sliders blend subway ambience, background chatter and lo-fi music over a train illustration by the artist mienar.

• Independent volume sliders for subway ambience, chatter and music.
• Light and dark themes that follow the system setting and remember the visitor’s choice.
• Responsive layout for phones, styled with Sass. Deployed on Netlify.
• Reached 610 new users in 12 months.

Co-op with Quinn Ha.

A study tool for recorded lectures: upload a video, watch it beside an AssemblyAI speech-to-text transcript, and (the plan) drag snippets onto your slides to export as a PDF. In 36 hours our team of four built upload, playback and transcription.

• Wrote the AssemblyAI calls that submit audio for transcription and fetch the transcript.
• Set up the repo and React app, merged the Express transcription server, and wrote the README and Devpost story.
• Teammates built the video upload page and upload server (Douglas Chen) and an Express server that polls AssemblyAI for the finished transcript (Aditya Bhadauriya).

Co-op with Douglas Chen, Aditya Bhadauriya, Fady Morcos.

A weekly workout planner that deliberately schedules rest days, designed to use Fitbit sleep and activity data to decide when to train and when to recover. Willie Pai and I built it in 36 hours at DeltaHacks 8.

• Wrote the Fitbit Web API client, which pulls 7-day sleep and activity logs and heart-rate data (a standalone Node script, not yet wired into the UI).
• Built the navigation bar and landing page with styled-components, plus the routes for the other pages.
• Merged the feature branches, deployed the React app to GitHub Pages, and wrote the README and Devpost story.
• Willie Pai built the weekly schedule view and the exercise GIFs.

Co-op with Willie Pai.