StarterChef

Your start to great cooking. A personalized cooking assistant for beginners: scan your kitchen to detect ingredients and equipment, get recipe suggestions ranked by what you already have, import recipes from a link, photo, text, or YouTube, and cook step-by-step with an AI sous-chef you can talk to and show your pan to.

• Live app: Click here
• Course: CS3216 Assignment 3 (Artificial Intelligence Application)

Team

Name	Matric no.	Contributions
Huang Kaijuan Joulene	A0299817E	Created the base app with core features and tech stack with supabase auth and posthog set up. Designed app logo and interactive feature introduction on the landing page and onboarding and profile page. Chose the AI models to use, wiring them in with Vercel AI SDK and writing initial prompts. Added per user AI rate limiting. Seeded base 6 recipes with TheMealDB. Added recipe import and core step by step cooking UX including the show camera and animated mascot. Added session memory to cooking sessions. Deployed the app to Vercel.
Cedric Tay	A0307676U	Implemented AI-specific scan review UX and human-in-the-loop controls. Developed the usage-based credit and pricing model, enhanced landing-page pricing, SEO and accessibility, fixed analytics tracking, and prepared Product Hunt launch materials.
Hoang Trung Hai	A0325468W	Worked on backend architecture, security, and AI infrastructure, including authentication and authorization, database RLS policies, private media storage, API validation, quota handling, and cross-user security. Additional work covered secure kitchen scanning, recipe recommendations, imports, AI-assisted recipe workflows, persistent cooking sessions, live voice and assistant integrations, telemetry, cleanup, testing, and production-readiness safeguards. Development followed phased vertical slices with explicit security, reliability, and testing gates throughout.
Angel	A0276261A	Implemented and debugged PostHog analytics across the complete user journey, including onboarding, kitchen-item saving, recipe recommendations, recipe selection, cooking-session starts, feedback submission, and session completion. Fixed cooking-flow issues and made feedback saving compatible with different database versions. Developed the AI evaluation datasets and runner for recipe recommendations and cooking assistance, including repeated model runs, request pacing, cost estimation, provider-error recording, safety criteria, and evaluation documentation. Used the evaluation results to improve and ground the cooking-assistant prompt. Also improved the landing page’s mobile responsiveness and pricing-table accessibility, updated milestone evidence, and prepared and verified Product Hunt launch materials.

Tech stack

• Next.js 16 (App Router, TypeScript) — responsive web app
• Tailwind CSS v4 — brand tokens in src/app/globals.css
• Vercel AI SDK — provider-agnostic LLM layer (src/lib/ai/), swappable via AI_PROVIDER env var
• Recipe verification routing — uses the selected provider end-to-end by default; set RECIPE_VERIFICATION_ROUTING=cross-provider to opt into the OpenAI/Gemini independent-verifier workflow (requires both keys)
• Supabase — auth, Postgres, storage (supabase/migrations/)
• Live voice — OpenAI Realtime first, with one pre-connection Gemini Live fallback
• Vitest + ESLint/Prettier + Husky/lint-staged + GitHub Actions

Local setup

cp .env.example .env.local   # fill in an AI provider key and Supabase keys
npm install
npm run dev                  # http://localhost:3000

Supabase checklist

1. Create a project at supabase.com.
2. In Project Settings → API Keys, copy the publishable key (starts with sb_publishable_) into NEXT_PUBLIC_SUPABASE_PUBLISHABLE_KEY, and the secret key (starts with sb_secret_) into SUPABASE_SECRET_KEY.
3. In Project Settings → Data API, enable the Data API and disable Automatically expose new tables.
4. Run the migrations in supabase/migrations/ in order. On an existing project, check the rollout docs in docs/ first — some phases need a preflight.
5. Load the starter recipe catalogue: run supabase/seed/recipes.sql in the SQL editor. The 6 recipes are sourced from TheMealDB for demo use; review and attribute them correctly before any public release.
6. In Authentication → URL Configuration, set:
   • Site URL: http://localhost:3000/today
   • Redirect URLs: http://localhost:3000/auth/callback and your production URL once deployed
7. Enable an auth provider (e.g. Google under Authentication → Providers) or keep email auth enabled.
8. (Optional) Disable Confirm email for password signups under Authentication → Providers → Email while developing — re-enable for production.

Useful scripts:

npm run lint          # ESLint
npm run typecheck     # tsc --noEmit
npm test              # Vitest unit tests
npm run eval          # AI eval suite (needs a provider API key)
npm run format        # Prettier

Repo layout

src/app/(marketing)/   landing page (SEO + OG image, interactive demos, pricing)
src/app/(app)/         authed app shell: today, kitchen, recipes, cook/[id], sessions, settings
src/app/login · onboarding · auth/  sign-in, first-run wizard, auth callback
src/app/robots.ts · sitemap.ts · manifest.ts · icon.svg   SEO + PWA metadata
src/app/api/ai/        AI endpoints (assistant, import-recipe, kitchen-voice,
                       realtime-sessions, suggest-recipes)
src/app/api/           app endpoints (cooking-sessions, recipe-drafts,
                       kitchen-scans, pantry-items, cron retention)
src/components/        shared UI — cards, cook assist, ChefBuddy, landing demos
src/hooks/             voice hooks: OpenAI Realtime + Gemini Live + shared assistant state
src/lib/ai/            provider abstraction, zod schemas, tools, voice telemetry
src/lib/supabase/      browser, server + admin clients, generated DB types
src/lib/data.ts        server-side query helpers (profiles, kitchen_items, recipes, sessions)
src/lib/credits.ts     plan + credit amounts — single source for pricing UI
src/lib/posthog/       analytics init + event tracking helper
src/instrumentation.ts PostHog AI observability via OpenTelemetry
src/proxy.ts           auth session refresh + route guard (Next.js proxy/middleware)
workflows/             durable workflows (recipe verification, session recap)
prompts/               versioned system prompts (cited in milestones writeup)
evals/                 eval datasets + runner (LLMOps milestone)
tests/                 Vitest unit + route-contract tests
e2e/                   Playwright end-to-end specs
scripts/               ops scripts (AI key smoke test, retention cron, media migration)
supabase/migrations/   database schema (RLS on every user table)
supabase/seed/         starter recipe catalogue (TheMealDB demo data)
docs/                  milestone writeups, rollout plans, credits model, launch assets
.devin/skills/         agent skills for project workflows

Resources used

• Vercel AI SDK — provider abstraction, structured outputs, tool calling
• Supabase SSR guide — auth client patterns
• CS3216 Assignment 3 spec — milestone requirements