A team of AI helpers that builds a healthy, affordable meal plan and keeps fixing it as your life changes.
HealthTypeScript · PLpgSQLGitHub @fernandomunoz-ruralOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
No updates detected yet. GitHub releases will appear here.
Owners of public repositories under a personal GitHub account can sign in to verify ownership and manage the introduction themselves. No administrator approval is needed.
Sign in, generate a verification file, and commit it to this repository’s default branch within 30 minutes. Files on other branches or forks do not count. Then check the file below to manage the project without administrator approval. Keep the file to retain your access.
If needed, you will go to sign-in. After signing in, return here and press your chosen verification button again.
Run /nomorevibe verify in your project folder to verify ownership and manage the introduction yourself.
Verify in one minuteAsk about other verification options
NutriPlan AI is a multi agent AI nutrition web app that combines LLM agent orchestration, computer vision, and mathematical optimization to generate, monitor, and autonomously adapt personalized meal plans for users with dietary conditions such as diabetes and hypertension.
Taken from the repository README
Loading the full README. The saved excerpt is shown below.
A team of AI helpers (Doctor, Chef, Budget, Monitor and an Orchestrator) that negotiates a healthy, affordable 7 day meal plan in front of you and keeps fixing it as your body responds.Read the full README ↗A team of AI helpers (Doctor, Chef, Budget, Monitor and an Orchestrator) that negotiates a healthy, affordable 7 day meal plan in front of you and keeps fixing it as your body responds.
Built for the Ensign BUILD hackathon. Real engineering under the hood: a hand written multi agent orchestrator on the Vercel AI SDK, a mixed integer program (HiGHS compiled to WebAssembly) whose shadow prices become plain language explanations, Kalman filter trends on weight, glucose and blood pressure, Bayesian metabolism (TDEE) updates, two vision models voting on food photos, and diabetes (ADA) and hypertension (DASH) rules with sources. Language models write the words; math produces every number.
, Mac 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 (or 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿; the repo has an 붿붿붿붿붿붿붿)., Mac 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 (or Xcode tools). / 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿). Open the repo and accept the recommended extensions.붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 붿붿붿붿붿붿 checks Node, Git and Docker, copies 붿붿붿붿붿붿붿붿붿붿붿붿붿 to 붿붿붿붿붿붿붿붿붿붿붿, runs 붿붿붿붿붿붿붿붿붿붿붿붿, installs the Playwright Chromium browser, creates 붿붿붿붿붿붿붿 and runs the typecheck. Use 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 to skip the browser download; 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 only reports.붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿`
Open http://localhost:3000 and click "Try as Alex". No API keys, no database, no network needed: the memory backend is seeded with the two demo users and the agents run in scripted mode. when you want live agents (붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿); restart 붿붿붿붿붿붿붿붿붿붿붿붿`.| Variable | Default | Options | What it does |
|---|---|---|---|
| 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿` | 붿붿붿붿붿붿` | 붿붿붿붿붿` | Frontend reads 붿붿붿붿붿붿붿붿붿붿붿붿붿붿` with a fake 600 ms delay; no API calls. |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿` | 붿붿붿붿붿붿붿` | 붿붿붿붿붿붿붿붿붿` | Memory: JSON under 붿붿붿붿붿붿붿, seeded. Supabase: Postgres with the keys in 붿붿붿붿붿붿붿붿붿붿붿붿붿. |
| 붿붿붿붿붿붿붿붿붿붿` | 붿붿붿붿붿붿` | 붿붿붿붿붿붿붿붿붿` | Local: cookie login as 붿붿붿붿붿, 붿붿붿붿붿붿 or any email. |
| 붿붿붿붿붿붿붿붿붿붿붿붿` | auto | 붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿 | Auto is live when a model key exists, else scripted (deterministic negotiation, offline). |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿` | 붿붿붿붿` | number | Pacing of the Agent Room. |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿` | 붿붿붿붿붿` | 붿붿붿붿붿붿` | Replay a saved negotiation if a live model fails. |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿` | blank | URL | Run Playwright against a deployed URL. |
Full list with comments in 붿붿붿붿붿붿붿붿붿붿붿붿붿; how to switch and recover in 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿.
| Command | What |
|---|---|
| 붿붿붿붿붿붿붿붿붿붿붿붿` | Next dev server on http://localhost:3000 |
붿붿붿붿붿붿붿붿붿붿붿붿붿붿 / 붿붿붿붿붿붿붿붿붿붿 | Production build and server |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿` | 붿붿붿붿붿붿붿붿붿붿붿붿붿` (strict) |
붿붿붿붿붿붿붿붿붿붿붿붿붿 / 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 | ESLint / Prettier |
| 붿붿붿붿붿붿붿붿붿` | Vitest unit tests for 붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿, 붿붿붿붿붿` |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿` | Playwright demo flow (projects 붿붿붿붿붿붿붿 1440x900 and 붿붿붿붿붿붿 390x844) |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿붿` | Agent negotiation evals |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿붿` | Regenerate 붿붿붿붿붿붿붿 and 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 from real math |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿` | Reset demo users Alex and Maria in the configured backend |
붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 / 붿붿붿붿붿붿붿붿 / 붿붿붿붿붿붿붿붿붿붿붿붿 / 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 | Food data imports |
붿붿붿붿붿붿붿붿붿붿붿붿붿붿 / 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 | First time setup / environment report |
붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 / 붿붿붿붿붿붿붿붿 / 붿붿붿붿붿붿붿붿붿` | Local Supabase via the CLI (Docker Desktop required) |
| Role | Person | Owns |
|---|---|---|
| 1 Design / Frontend | 붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿, 붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿` | |
| 2 AI Agents | 붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 | |
| 3 Backend / Data | 붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿붿붿붿, all other 붿붿붿붿붿붿붿붿붿 routes | |
| 4 Math | 붿붿붿붿붿붿` | |
| 5 Lead / Integration | 붿붿붿붿붿붿붿, root config, 붿붿붿붿붿붿붿붿붿붿 files, 붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿, 붿붿붿붿붿붿붿붿붿붿붿, merges, deploy |
Only the Lead edits 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿; requests go through the group chat. Briefs per role live in 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿; the master spec is 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿`.
Screens get data only through 붿붿붿붿붿붿붿붿붿붿붿붿 (mock switch inside). It calls route handlers under 붿붿붿붿붿붿붿붿붿, which authenticate with 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 and persist through the 붿붿붿붿붿붿 interface in 붿붿붿붿붿붿붿붿 (memory or Supabase). Plan generation runs the Orchestrator in 붿붿붿붿붿붿붿붿: Doctor builds hard constraints from 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿, Chef proposes candidate meals mapped to real foods, Budget prices them, and the optimizer in 붿붿붿붿붿붿 solves a week as a MILP; the negotiation streams to the Agent Room over Server Sent Events and is stored for replay. Readings and photo logs go through the Monitor, which uses Kalman trends and a Bayesian TDEE update to replan on its own. Diagrams in 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿`.
| Problem | Fix |
|---|---|
| Windows: Prettier or ESLint complain about CRLF, or git shows every file modified | 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 forces LF. Run 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 in the repo, then 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 once. VS Code setting 붿붿붿붿붿붿붿붿붿붿 is already 붿붿붿`. |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿` | Windows: 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 then 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿. Mac: 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿. Or 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿. |
붿붿붿붿붿붿 WebAssembly fails to load (붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 or 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿`) | 붿붿붿붿붿붿 is listed in 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿; make sure you did not import it from a client component. Delete 붿붿붿붿붿붿 and restart. The optimizer falls back to 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 automatically. |
| Docker Desktop not running | Only matters for 붿붿붿붿붿붿붿붿붿붿붿붿붿 and 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿. Start Docker Desktop, or stay on 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿`. |
| 붿붿붿붿붿붿붿붿붿붿붿붿붿붿` says Node is too old | Install Node 20+ (붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 or 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿) and reopen the terminal. |
| Playwright cannot find Chromium | 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 (CI uses 붿붿붿붿붿붿붿붿붿붿붿붿). |
| 붿붿붿붿붿붿붿붿붿붿붿` changes have no effect | Next reads env at startup. Stop and restart 붿붿붿붿붿붿붿붿붿붿붿붿`. |
| Agent Room shows nothing | No keys and 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 forced: unset it (auto picks scripted). Check the dev console for 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿. |
| PowerShell: 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿` fails | Use 붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿붿 or put the variable in 붿붿붿붿붿붿붿붿붿붿붿. |
MIT, NutriPlan AI team. See 붿붿붿붿붿붿붿붿`. NutriPlan AI applies published nutrition guidelines and is not medical advice.
To run against the team database (same data for everyone, real auth, photo storage) follow docs/supabase-setup.md. You need the shared .env.local from the Lead; without it the app runs on the local memory store.