AI Chat

The AI Chat is the primary interface for research and discovery. Advisors interact with a natural language assistant that understands intent, extracts parameters, and fans out to vertical-specific adapters.

Interaction Flow

1. Initial Request — Advisor types a natural language query (e.g., "Find sushi restaurants in Tokyo for 2 people")
2. Intent Detection — Keyword matching classifies the request (dining, hotels, flights, experiences, transportation, DMCs)
3. Parameter Extraction — Regex extracts location, dates, party size, cuisine preferences
4. Data Source Selection — Advisor selects which data sources to search (Resy, OpenTable, Yelp, etc.)
5. Search Execution — Console sends POST /api/v1/search to the Orchestrator
6. Results Display — Results render as cards with source badges, time-slot availability, and pricing
7. Hand-off — Advisor clicks "Hand off to Atlantes" to create a Quote__c or Booking__c

NLP Pipeline

Intent Detection

Keyword matching classifies requests into verticals:

Vertical	Keywords
Dining	restaurant, dinner, lunch, sushi, reservation, book a table
Hotels	hotel, stay, room, accommodation, nights
Flights	flight, fly, airplane, airline, airport
Experiences	tour, activity, experience, attraction, thing to do
Transportation	car, transfer, taxi, ride, driver
DMCs	DMC, destination management, local operator

Fallback to LLM if keyword matching fails.

Parameter Extraction

Regex patterns extract structured parameters from natural language:

• Location: in [location], at [location], near [location]
• Dates: next [day], this [day], tomorrow, MM/DD/YYYY
• Party size: for [N] people, party of [N], [N] guests
• Cuisine: sushi, italian, mexican, etc.

Data Sources

Advisors toggle data sources per search:

Source	Type	Notes
Resy Direct	API	Real-time availability
Resy Parse	Scraping	Fallback when API unavailable
Yelp	API	Reviews and ratings
OpenTable	API	Reservations
Tock	API	Experiences
Toast Tab	API	Restaurant data
TheFork	API	European restaurants

Result Display

Results render as cards with:

• Restaurant/venue name and description
• Source badge (which data source provided the data)
• Time-slot availability (green badges for available slots)
• Price estimate
• Hand-off button

Chat History

Sessions persist in IndexedDB with full restoration:

• Messages (user and assistant)
• Context references (member/trip records)
• Detected intent and location
• Search filters and results
• Data source selections

History displays in a sidebar with RECENT/EARLIER grouping and debounced auto-save (2 second delay).

Preset Routines

Six preset routines on the AI landing page:

1. Search Dining
2. Book Flight
3. Draft Itinerary
4. Find Hotels
5. Research Destination
6. Compose Email