What is Curator and Why it Matters?

The rapid adoption of Generative AI has fundamentally shifted how travelers plan, research, and experience their journeys. To capitalize on this, building an authoritative FAQ knowledge base within the Bonafide platform is essential for training and grounding Large Language Models (LLMs) to provide accurate, property-specific answers.

This overview outlines the strategy for the creation, curation, and authentication of this specialized dataset. While establishing a ground-truth reference for AI often requires significant manual effort, this workflow leverages the Bonafide UI – Curator Module to streamline the process, transforming a daunting data entry task into a highly targeted and efficient operational workflow.

1. Creation: Sourcing Real-World Traveler Intent

The foundation of this knowledge base is rooted in actual user behavior rather than theoretical assumptions.

• Prompt Origin: The prompt set is derived from the most frequent, most relevant, real-world queries travelers are currently asking generative AI platforms.
• Relevance & Scope: By focusing on essential property aspects and amenities (e.g., specific amenity requests, site features, services, local area recommendations, policy inquiries), the intent is to establish prompt responses with the highest relevance to the question at hand.
• Structuring the Data: These real-world prompts are mapped to specific properties, and include several key parameters allowing ‘curators’ to effectively and efficiently curate initial responses garnered from in-depth property site ‘crawls’ and potentially from LLM consensus, when applicable. 

1. Curation: Streamlining Effort via the Bonafide UI

Curating a comprehensive dataset across multiple properties and feature sets is traditionally bottlenecked by data fatigue and disorganization. The Bonafide Curator Module is designed to expedite the curation process into a highly visible, manageable and prioritized pipeline.

Instead of navigating endless spreadsheets, users are empowered to filter and address content across key strategic data points:

• Feature Types & Feature Names: Curators can isolate specific categories (e.g., filtering down to just "Dining" or "Accessibility Features"), allowing for focused, batch-oriented data entry.
• Property Names: Enables portfolio-wide management by allowing users to toggle between individual locations or view gaps across a broader group of properties.
• Priority Ranking: Not all FAQs are created equal. The UI surfaces high-impact, frequently asked queries, ensuring that the most critical data is established before moving to edge cases.
• Blank Response Entries: The system highlights missing data fields. By isolating blank fields, curators can rapidly close knowledge gaps without hunting through completed work.

1. Authentication: Establishing the "Ground Truth"

For an LLM to be reliable, the data it ingests must be absolutely authoritative. Hallucinations occur when models lack clear, verified information.

• Data Initialization: Baseline responses are established at the outset of an engagement.  Responses are identified and translated into baseline summaries derived from specific source elements on property website pages.  This baseline training dataset (incl. Official Responses and Official Links) is used as a baseline for the Knowledge Base and provides a tremendous head start toward authentication and curation efforts.
•  Data Verification: Once responses are drafted/updated in the Bonafide UI, they also require an additional authentication step to ensure validity. This ensures that the information provided represents the official, current stance of the property.  When a response is not ready for Orchestration, a simple assignment change (to “Unverified”) will flag for review and will also not be socialized to LLMs (in an Orchestration).
• LLM Readiness: The authenticated data is then structured in a format optimized for LLM ingestion (.md and .html formats within the Bonafide FAQ Package).
• Continuous Maintenance: Because the Bonafide UI surfaces blank entries, and with the priority filter in place, maintaining the integrity of this "Knowledge Base" becomes a continuous, sustainable process rather than a massive annual overhaul.  Future platform enhancements are intended to raise a flag when specific responses require attention.
• Context and Importance: Additional references within the Bonafide platform can also provide guidance on which Feature Types to focus on and what questions, in particular, need to be Curated.

Summary

By establishing a prompt set based on real-world traveler prompts, subsequently crawling sites for relevant content and then utilizing the filtering and prioritization capabilities of the Bonafide Curator Module, the effort required to build an effective knowledge base is significantly reduced. The result of these exercises is a curated, authenticated, and highly structured data asset that empowers LLMs to act as confident, accurate digital concierges.