DAHAB• دهب Guest App · Product Specification
10 Guest App · Module

AI Concierge Chat

A chat assistant, opened from anywhere in the guest app, that answers practical questions about Dahab (visibility and currents, wind, safety, transport, what to pack, culture) in the guest's language, and surfaces bookable listings and guide articles inline as tappable cards.

Module
AI Concierge chatbot (drawer / bottom sheet)
Apps
Desktop Web (1440), Tablet Web (834), Mobile Web, Native App (390)
Figma
Not in the Figma file yet. The screens exist in Stitch ("AI Concierge — Chat Drawer Empty State" and "AI Concierge — Active Conversation State", 4 breakpoints each) but were not added to page New. Screenshots will be added once they are in Figma.
SRS
§3.10 FR-10.1, FR-10.2 (FR-10.3 Later), §3.11 FR-11.2, §5.3 LLM API, §2.6 LLM provider dependency
Screens
1 component · 2 states · 8 Stitch screens

1. Module summary

A chat assistant, opened from anywhere in the guest app, that answers practical questions about Dahab (visibility and currents, wind, safety, transport, what to pack, culture) in the guest's language, and surfaces bookable listings and guide articles inline as tappable cards. It is a slide-over drawer (desktop/tablet) or bottom sheet (mobile/native), not a full page, so the guest keeps their browsing context.

2. Business requirements covered (SRS)

ID Requirement Priority MVP Where
FR-10.1 Chat with an AI assistant in their language about practical Dahab questions High MVP ✓ Empty state + conversation
FR-10.2 Chatbot surfaces relevant properties/activities/guide articles inline Medium MVP ✓ Inline listing cards in answers
FR-10.3 Escalate to human support Medium Later Not designed (safety disclaimer only)
FR-11.2 Popular / similar suggestions for new users Medium MVP ✓ Suggested starter questions
§5.3 LLM/AI API — MVP ✓ Backend

3. Screen inventory

# State Breakpoints (Stitch) Presentation
4.1 Chat drawer — Empty state All 4 Desktop: right slide-over, full height, 460 px wide, blurred page behind · Tablet: 420 px slide-over with scrim · Mobile Web: bottom sheet with drag handle · Native: iOS-style modal sheet
4.2 Chat drawer — Active conversation All 4 Same containers; desktop drawer 480 px

4. Screen specifications

4.1 Empty state

Screenshots pending — not in Figma.

Element Content (Stitch spec) Behaviour / rule
Launcher Entry point to open the concierge (floating button / header icon) Not specified in the designs — see §7 G2.
Header Title, language quick switch عربي, close ✕ (native: grabber bar + close) Close keeps the conversation in memory for the session.
Greeting "Marhaban! I'm your Dahab Coastal Guide" / "Ahlan! What can I guide you to?" — capabilities: tides, Blue Hole visibility, wind, Bedouin safaris, transport Greeting in the UI language.
Live conditions pills "Laguna: 14 kts NNE · High tide at 16:40 · Blue Hole: 35m+ visibility · Chamber standby 24/7" Same data source as module 02.
Suggested starter questions (4 cards) 🌊 "What's the visibility and drift like at Blue Hole today?" · 🤿 "Find me a beginner-friendly dive site near Eel Garden or Lighthouse" · 🎒 "What should I pack for an overnight Bedouin desert safari in Wadi Ghnai?" · ☕ "Where can I enjoy authentic campfire habak tea at sunset with locals?" One tap sends the question.
Input bar Placeholder "Ask about dive spots, wind, treks, or stays…", voice dictation, terracotta send button Send disabled when empty.
Disclaimer "Data verified by local Bedouin guides & CDWS certified dive masters. For marine emergencies call VHF Ch 16." Always visible.

4.2 Active conversation

Screenshots pending — not in Figma.

Element Content (Stitch spec) Behaviour / rule
Guest message "Find me a beginner-friendly dive site near Eel Garden or Lighthouse" · 10:24 AM Timestamp; delivered state.
AI answer Local greeting, comparison (Lighthouse Reef sheltered pebble entry & zero drift vs Eel Garden surge today), live metrics (35m+ visibility, calm, 26°C) Answer in guest's language; grounded in platform data (listings, guide content, conditions).
Inline activity card "Lighthouse Reef Guided Beginner Scuba & Skills Refresher" — instructor Youssef Ben-Ammar, CDWS badge, 12 m, gear included, ★ 4.95, 1,200 EGP/person — Book Slot / View Session Card → Activity Detail / Session Picker (05/07). Only real, approved listings.
Inline stay card "Dar Tarabin Coastal Ecolodge" — host Sheikh Salama, "4 minutes from Lighthouse dive center", ★ 4.96, 2,400 EGP/night — Explore Stay Card → Property Detail (05). Native shows cards as a horizontal carousel.
Follow-up chips Suggested next questions One tap sends.
Input "Ask follow-up or request booking…", voice, photo attachment, send Photo attachment — see §7 G4.
Typing / loading state AI "typing" indicator Required while waiting for the LLM.

Step-by-step

# Guest does System does
1 Opens the concierge Opens drawer/sheet over the current page; shows greeting, conditions, 4 starter questions.
2 Taps a starter or types a question Shows the message; shows typing indicator; calls the LLM with context (language, current page/listing, conditions, catalogue search).
3 — Streams the answer; attaches up to N listing/article cards from the catalogue (FR-10.2).
4 Taps a card CTA Navigates to detail/booking; drawer closes (desktop may keep it open).
5 Asks a follow-up Keeps conversation context.
6 Asks something unsafe/unknown Answers with safety guidance and emergency contacts; (escalation to human is Later — FR-10.3).
7 Closes the drawer Conversation kept for the session (history persistence — decide).

5. Business rules

  1. Answers must only recommend approved, active listings and published guide content; prices and availability come from the platform, never generated by the model.
  2. Answer language = UI language (EN/AR), or the language the guest writes in.
  3. Safety: medical/diving-emergency questions always include the hyperbaric chamber and VHF 16 contacts and a "not a substitute for a certified guide" note.
  4. Store chat sessions (entity AI Chat Session, SRS §6) for quality review; retention and privacy rules per NFR-9.
  5. Rate limits per user/device to control LLM cost.
  6. Works logged-out (read-only answers); booking still requires login.

6. Story candidates for Jira

Epic: GUEST-AI-CONCIERGE

Key idea Story Acceptance criteria
AI-1 As a guest, I want to open a chat assistant from any page. Drawer (desktop/tablet) / bottom sheet (mobile/native); close keeps context; launcher defined.
AI-2 As a guest, I want suggested questions to start. 4 starter cards; tap sends.
AI-3 As a guest, I want answers about Dahab in my language. EN/AR answers; typing indicator; streamed response; error state if LLM fails.
AI-4 As a guest, I want the assistant to show bookable stays/activities and articles as cards. Cards link to detail/booking; only live catalogue items; correct price.
AI-5 As a guest, I want safety-critical answers to include emergency contacts. Chamber + VHF 16 shown for safety topics; disclaimer visible.
AI-6 (tech) LLM integration with catalogue retrieval, logging and rate limiting. Prompt includes language + retrieved listings/articles; sessions logged; per-user limits.

7. Gaps & open questions

# Finding Suggested decision
G1 Concierge screens are not in the Figma file (only in Stitch). Copy the 8 Stitch screens into Figma page New (section "AI Concierge").
G2 Launcher (where the guest opens the chat) is not designed on any screen. Add floating button / header icon to the shell.
G3 Error, offline and "can't answer" states not designed. Add sibling states.
G4 Photo attachment & voice dictation in the input are not in the SRS. Phase 2 or confirm scope.
G5 Human escalation (FR-10.3) is Later — define fallback (WhatsApp desk link?). Show "Chat with the Dahab desk on WhatsApp" link.
G6 Conversation history persistence across sessions not specified. Decide (session only for MVP).

8. Out of scope for MVP

  • Human escalation (FR-10.3), AI trip planner (FR-9, Phase 2), personalised recommendations (FR-11.1).