Final Capstone —
ReceptionistAI SaaS

• Landing → sign up → connect Twilio number → upload FAQ → clone voice → go live.
• Owner sees call logs, transcripts, bookings, and analytics.
• Stripe subscriptions ($99 / $299 / $999 tiers).
• Google Calendar 2-way sync.

Browser (TanStack Start dashboard)
       │
       ▼
 App backend (server functions)  ──▶  Supabase (Postgres + Auth + Storage)
       ├──▶ Twilio (voice + numbers)
       ├──▶ ElevenLabs (TTS + cloning)
       ├──▶ Claude (reasoning)
       ├──▶ Deepgram (streaming STT)
       ├──▶ Google Calendar
       └──▶ Stripe (billing)

create table tenants (
  id uuid primary key default gen_random_uuid(),
  name text, plan text check (plan in ('starter','growth','scale')),
  stripe_customer_id text, created_at timestamptz default now()
);
create table calls (
  id uuid primary key default gen_random_uuid(),
  tenant uuid references tenants, from_e164 text, duration_s int,
  transcript jsonb, intent text, outcome text, cost_usd numeric
);
create table bookings (
  id uuid primary key default gen_random_uuid(),
  tenant uuid references tenants, customer_name text, phone text,
  slot timestamptz, gcal_event_id text, status text
);

• Hero: "Never miss a call again. $99/mo."
• 60-second demo call widget.
• 3-tier pricing, testimonials, FAQ, big CTA.

Supabase Auth (email + Google). RLS on every table keyed by tenant. Roles in a separate user_roles table.

• KPI row: calls today, bookings today, avg duration, resolution rate.
• Live call feed with expandable transcripts.

Persona picker, sliders for stability/similarity/style, preview button that hits ElevenLabs live.

Paginated table + transcript viewer + audio playback.

Callers dedup'd by phone number. Timeline of every call + booking + note per contact.

tools: [{
  name: "bookAppointment",
  input_schema: { type: "object",
    properties: { customer_name:{type:"string"}, phone:{type:"string"},
      slot_iso:{type:"string",format:"date-time"} },
    required:["customer_name","phone","slot_iso"] },
}]

OAuth per tenant. Store refresh token. Insert events on booking; delete on cancel.

1. Sub-account per tenant.
2. Buy a number in the tenant's area code.
3. Point voice webhook at /api/public/twilio/incoming.

Stream with eleven_turbo_v2_5. Store voice_id per tenant.

System prompt composed from tenant name + business hours + FAQ chunks + persona tone. Max 200 tokens per reply.

• Upload PDFs/URLs → chunk → embed → store in pgvector.
• Every turn: select top 5 chunks by cosine, splice into system prompt.

In-app recorder → 90-second script → upload to ElevenLabs IVC.

• Calls/day, resolution %, booking rate, avg cost/call.
• Intent breakdown pie chart.
• Latency histogram (goal p95 < 1.8s).

const session = await stripe.checkout.sessions.create({
  mode: "subscription",
  line_items: [{ price: PRICE_IDS[plan], quantity: 1 }],
});

• Dashboard → Cloudflare/Vercel.
• Voice agent → Railway with 2 always-warm instances.
• Sentry + BetterStack + Grafana.

1. Terms + privacy + call-recording disclosure.
2. 10 successful synthetic calls.
3. Stripe live-mode swap.
4. Status page + onboarding email + Loom.

• Landing SEO — 5 pages per vertical.
• Cold outbound to 100 salons + 100 clinics.
• Product Hunt launch on a Tuesday.
• YouTube demo + weekly changelog.