Gamma

Coaching client management. A Next.js app that turns coach documents into a confirmed "active protocol", ingests real nutrition/health data from Health Connect via an Android companion app, and closes the loop with a compliance dashboard, an AI weekly analysis, and a generated coach check-in.

Live demo: https://show-prep-gamma.vercel.app — seeded with sample coaching data; log in with the demo passcode.

Monorepo layout (pnpm workspaces):

• / — Next.js 16 (App Router) + TypeScript + Drizzle + PostgreSQL web app
• /mobile — Expo (React Native) Health Connect companion for Android (its README)

Features

• Documents — upload coach PDFs/txt or paste emails. Claude (claude-opus-4-8) extracts structured prescriptions (calories, P/C/F, effective date, cardio plan); you review and confirm before one becomes the active protocol. Program rules/guidelines are just another document category.
• Dashboard — countdown to target date, current weight vs target weight with a 7-day trend, daily macro compliance vs the active protocol, and an AI-written plain-language weekly analysis.
• Check-in — answers your coach's exact template (stored as app config): data-backed questions (macro adherence, bodyweight, water ≥3L, sleep ≥7h, workouts ≥3/wk & cardio, next target date) are pre-filled from ingested data; subjective ones (waist, strength, digestion, change requests) are manual fields saved to check_ins. Thresholds live in the weekly_targets table (set per client by the coach, on the client's page — Edit). Output: a copyable/mailable filled-in template.
• Client management (coach) — the Clients page lists clients; each client's page is read-only until Edit, which reveals name, email, target name/date/weight, program type, nutrition target, and weekly check-in thresholds for that client. A coach's own Settings page is just "Add a client" — there are no coach-level goals.
• Doc chat — RAG over your uploads (Voyage AI voyage-4 embeddings, 1024-dim, pgvector), with source citations.
• Log data — manual fallback for weight, sleep, water, and daily activity (/log, POST /api/manual-entry): a client logs their own days, a coach picks a client. Saved with source: "manual"; saving a day again replaces that day's manual entry. Macros/nutrition are not logged here.
• Ingest API — /api/ingest/{nutrition|weight|hydration|sleep|exercise|activity}, Zod-validated, bearer-token-gated, idempotent (upserts on Health Connect record UIDs).

Quick start

pnpm install
cp .env.example .env.local   # add ANTHROPIC_API_KEY + VOYAGE_API_KEY for AI features
pnpm seed                    # optional: 35 days of demo data (run with dev server stopped)
pnpm dev

No database setup needed: without DATABASE_URL the app runs an embedded PGlite (WASM Postgres, pgvector included) at .data/pglite. Point DATABASE_URL at a real Postgres (with the pgvector extension) to use one — same Drizzle schema/migrations either way.

PGlite is single-process: stop the dev server before running pnpm seed, and vice versa.

Deploying (Vercel + Neon)

1. Create a Postgres database (Neon, or Vercel → Storage → Postgres) and copy its pooled connection string. Migrations and the vector extension are applied automatically on first boot — no manual setup.

2. Import the repo at vercel.com/new and set env vars: DATABASE_URL, ANTHROPIC_API_KEY, VOYAGE_API_KEY, INGEST_API_KEY (required in prod — the ingest API is public), SESSION_SECRET, and for a public demo SEED_COACH_PASSCODE/SEED_CLIENT_PASSCODE (see "Demo account seeding" in .env.example — sets the seeded demo account's passcode; there's no login-page shortcut, sign in with it like any other account).

3. Seed the deployed database with sample data + AI content from your machine:

   DATABASE_URL="<neon-pooled-url>" SEED_AI=1 pnpm seed
   

AI routes run with maxDuration = 300; next.config.ts bundles the migration SQL into the serverless output.

Commands

Command	What
pnpm dev / pnpm build / pnpm start	Next.js
pnpm seed	demo data (35 days, active + pending protocol)
pnpm test	unit tests (Vitest)
pnpm test:e2e	e2e smoke tests (Playwright)
pnpm typecheck	tsc --noEmit
pnpm db:generate	regenerate Drizzle migrations after schema changes
pnpm backfill-accounts	one-time: create coach/client accounts, assign existing data (see the script's header comment)

Configuration

All optional — see .env.example. INGEST_API_KEY protects the ingest API (the companion app sends it as a bearer token). SESSION_SECRET enables the per-account login gate (each account has its own passcode, see scripts/backfill-accounts.ts); leave unset for none.

Day bucketing is done in America/Los_Angeles (configurable per client — a coach sets it on the client's page via Edit; a client sets their own in Settings): timestamps are stored as UTC and each row also stores its local_date, computed at ingest.

Data flow

MyFitnessPal ─┐
              ├→ Health Connect (on-device) → mobile companion (batched POST)
Samsung Health┘        → /api/ingest/* → Postgres → dashboard / check-ins

Coach docs → upload → Claude extraction → pending protocol → user confirms
          → active protocol (compliance baseline)
          → chunked + embedded (voyage-4) → doc chat (RAG)