LLM Integration (Speculative)
Relevant source files
The following files were used as context for generating this wiki page:
The Grapevine platform is described as an "AI-native platform" centered around a conversational AI travel assistant. This assistant leverages Large Language Models (LLMs) to provide 24/7, multi-channel support to business travelers. The core proposition is to combine the intelligence of public LLMs with the security, context, and compliance requirements of an enterprise environment, creating a "secure walled garden" for all interactions. This system aims to drive in-policy bookings, enhance duty of care, and automate post-booking communications.
This document outlines the architecture and features of the LLM integration as described in the project's public-facing website content. The information is based on marketing descriptions and high-level feature summaries, as no backend implementation files are available for analysis.
Core Platform Architecture
The LLM integration is part of the "Grapevine Core Platform," which is described as a vertically integrated system. The platform is structured in five distinct layers that work together to power the AI assistant.
Sources: app/product/page.tsx:34-39
The following diagram illustrates the five layers of the Grapevine Core Platform.
Sources: app/product/page.tsx:40-69
Architectural Layers
| Layer | Description | Source Files |
|---|---|---|
| Data Layer | Handles real-time data ingestion from sources like GDS, OBT, and mid/back-office feeds. | app/product/page.tsx:43-45 |
| Context Layer | Manages traveller-specific information, including profiles, preferences, and historical booking data. | app/product/page.tsx:48-50 |
| Assistant Layer | The core of the LLM integration, providing secure, multi-channel conversational AI. | app/product/page.tsx:53-55 |
| Marketplace Layer | Contains bookable inventory for hotels and other ancillary services. | app/product/page.tsx:58-60 |
| Insight Layer | Provides dashboards for tracking key metrics like spend, compliance, and traveller engagement. | app/product/page.tsx:63-65 |
The Assistant Layer
The Assistant Layer is the central component for LLM functionality. It is described as a "secure, multi-channel conversational AI" that provides the functionality of public LLMs within a controlled enterprise environment.
Sources: app/product/page.tsx:53-55
Secure Walled Garden Concept
A key feature of the assistant is the "secure walled garden." This concept is designed to prevent data leakage by combining the power of public LLMs with the specific context, control, and compliance needs of the enterprise. It ensures that sensitive corporate and traveller data does not get exposed to public models directly.
Sources: app/product/page.tsx:37-39, app/product/page.tsx:115-117
This diagram shows the conceptual data flow of the secure walled garden.
Functionality and Capabilities
The AI assistant offers several capabilities aimed at supporting the traveler throughout their journey.
| Capability | Description | Source Files |
|---|---|---|
| Multi-channel & Multi-modal | Accessible via chat or voice on SMS, WhatsApp, Slack, Teams, and email. | app/product/page.tsx:112-114 |
| System Integration | Fully integrated with booking and expense systems for seamless updates. | app/product/page.tsx:118-120 |
| Preference Learning | Learns traveller preferences over time (e.g., "likes sushi restaurants") to provide smarter recommendations. | app/product/page.tsx:121-123 |
| 24/7 Assistance | Provides round-the-clock trip information and assistance, regardless of the traveler's location. | app/page.tsx:81-82 |
| Secure AI Help | Delivers AI-powered assistance within secure and approved corporate channels. | app/page.tsx:87-88 |
Data Flow and Personalization
The AI assistant's effectiveness relies on its ability to combine data from various sources to deliver personalized and context-aware responses. The Data and Context layers feed information into the Assistant Layer to enrich interactions.
Sources: app/product/page.tsx:34-50
Data Sources for Personalization
The assistant leverages several data points to tailor its engagement:
- Itinerary Gaps: Identifies missing components like hotels, transfers, or parking.
- Traveller & Corporate Context: Understands who the traveller is, their employer, their destination, and their policy tier.
- Historical Data: Uses past booking data from the traveller and their colleagues to surface relevant options.
- Negotiated Rates: Highlights hotels with corporate negotiated rates.
Sources: app/product/page.tsx:79-85, app/product/page.tsx:96-98
Interaction Sequence
The following sequence diagram illustrates a typical interaction where the system detects an itinerary gap and engages the traveler.
Sources: app/product/page.tsx:79-99
Use Cases
The LLM-powered assistant is positioned to address several key use cases in business travel for both Travel Management Companies (TMCs) and corporate clients.
For TMCs
- Automate Post-Booking Follow-up: Reduces manual consultant workload by automatically engaging travelers.
- Capture Ancillary Revenue: Proactively prompts travelers to book hotels and other ancillaries, increasing revenue per trip.
- White-Label AI Assistant: Enhances the TMC's brand by offering a modern, AI-powered support channel.
Sources: app/tmcs/page.tsx:55-60, app/tmcs/page.tsx:68-73
For Corporates
- Improve Compliance: Guides travelers to make in-policy decisions without restricting choice.
- Strengthen Duty of Care: Maintains visibility across all trip components by keeping bookings within the managed program.
- Enhance Employee Experience: Provides travelers with a modern, 24/7 support tool.
Sources: app/corporate/page.tsx:60-75
Conclusion
The speculative integration of LLMs is a cornerstone of the Grapevine platform's identity. It is marketed as an "AI-native assistant" that moves beyond simple automation to provide intelligent, contextual, and secure conversational support for business travelers. By creating a "secure walled garden," the system aims to offer the convenience of modern AI tools while mitigating the risks of data leakage and non-compliant behavior associated with public AI. The primary goals are to increase ancillary revenue for TMCs, improve policy compliance and duty of care for corporates, and deliver a superior, personalized experience for the end traveler.
Sources: app/product/page.tsx:29-32, app/about/page.tsx:20-22, lib/seo.ts:111-115