1. The one-line

A customer-facing AI concierge for a Japanese izakaya. Three languages, auto-detected. Menu, allergens, sake pairings, reservations — all answered inside one chat panel, all grounded in what the venue actually knows. Built in two days for a Toronto pre-opening, end to end.

I came to this work as a former heavy-industry QA welder in nuclear power plant maintenance, and later as a Vancouver front-of-house manager. I bring a zero-defect mindset to bilingual F&B operations. A guest who asks about a dress code, a peanut allergy, or a private room for six on Friday is asking three completely different questions — but they expect one calm voice. This concierge is built to be that voice.

https://www.loom.com/share/904e3b32958441808d047092798896c4

90 seconds. Watch a multilingual guest browse the menu, run an allergen check, and book a private room — without ever leaving the chat panel.

HINOKAMI Toronto

Or: try it yourself.


2. Why this exists

This is the fourth project in a deliberate Portfolio sequence aimed at Japanese F&B groups operating in North America. The first three live behind the line — staff, service floor, area management. This one sits in front of the guest.

Portfolio 01 (SOP Generator) Portfolio 02 (Allergen Decision) Portfolio 03 (Daily Digest) Portfolio 07 (this)
Stakeholder Back-office / staff Service-floor server Area manager Guest
Use case Bilingual SOP generation Voice → allergen verdict Multi-location daily digest Browse → allergen → reserve
Cadence On-demand Realtime (≤ 1 second) Scheduled (daily 06:00) Realtime (chat)
AI weight Heavy (Sonnet 4.6) Light (Haiku, intent only) Moderate (Haiku, digest) Moderate (Haiku, conversational)
Cost / interaction ~$0.11 / generation ~$0.0003 / query ~$0.004 / digest ~$0.001 / chat turn (cached)
Sustainable SaaS price $300+/month $30–150/month per shop $50–200/month per chain $50–200/month per venue
Form factor Google Docs add-on Web app on server's phone Slack bot + Notion DB Customer-facing web site + chat

A venue that adopts all four inherits one allergen matrix as the source of truth, gets back-office documentation / service-floor decisions / daily ops oversight from the same canonical workflow logic — and now puts a guest-facing surface on the same operational discipline. Four tiers, same zero-defect spine.

The guest-facing problem is not "talk to a customer." Generic off-the-shelf chatbots already do that. The problem is: answer the way an experienced operator would. A guest asking about a peanut allergy expects the answer to be grounded in the same matrix the kitchen actually uses. A guest asking about a private room for six on Friday expects to know it seats four-to-eight, costs an extra $30/person room fee, and has a 72-hour cancellation policy — not the 24-hour policy that applies to counter seats. Those distinctions live in an operator's head. The concierge encodes them.

The reference build is HINOKAMI Toronto, a fictional pre-opening landing for a Spring 2027 izakaya on King West. Every operational rule it answers is grounded in real restaurant practice.


3. The demo, end to end

A single guest journey, narrated step by step. The Loom demo shows this exact path.

Browse — the Hero, the menu, the room

The guest lands on the homepage. Hero copy reads "A modern izakaya rooted in robata fire, the season's sashimi, and the quiet ritual of pouring sake for a friend." Four counters introduce themselves below, each with a counter-theme photograph and a 3-5 sentence description: Robata (straw flame + binchotan), the Counter (sashimi to order), Donabe & Sake (clay-pot rice + 24 sake labels), Drinks (highballs, wine, soft). A floating amber button sits in the lower right: Try the AI Concierge ↓.