GeomeeGo has established itself as a prominent platform for flight and hotel bookings, utilizing real-time inventory management through partnerships with Duffel for live flight content and TravelgateX for hotel distribution. The platform facilitates a straightforward process where users can search, compare, and book their travel arrangements in three easy steps, ensuring instant confirmation and encrypted transactions.
The upcoming focus for GeomeeGo is to enhance its functionality beyond conventional booking systems, aiming to transform the platform into a comprehensive operating system for today's travelers. This operating system design suggests a more sophisticated system than mere search functionalities linked with a chat feature. It emphasizes the platform's ability to maintain a user's objectives, gather necessary information, act on the user’s behalf, and provide transparency in its operations. This approach elevates the standard for AI travel tools significantly and is embodied in what GeomeeGo refers to as the Structured Cognitive Loop (SCL).
The challenges faced by current AI travel agents are notable. A common trait shared among these modern AI booking assistants is their architecture, where a single large language model performs multiple tasks: processing user requests, retaining conversation context, determining search outcomes, and completing bookings in an uninterrupted text flow. While such systems can produce impressive results during demonstrations, they tend to falter once conversations grow longer, leading to the loss of critical early details and leading to ambiguity in reasoning. Errors can emerge, such as booking the wrong fare class or incorrect travel dates, and when issues arise, users are left without clarity on the rationale behind these decisions. This situation is particularly problematic in travel, as bookings involve real monetary transactions subject to significant cancellation fees, making an acceptable margin of error untenable.
The introduction of the Structured Cognitive Loop will fundamentally change how these systems operate. This model delineates functions typically bundled into one, thereby distributing judgment across specialized components, each assigned a specific task. The retrieval phase establishes a set pool of information at the beginning of a query, avoiding ad hoc data extraction. In the cognition phase, language models generate action proposals rather than making decisions directly. The control aspect integrates deterministic programming that validates proposed actions against predefined criteria, preventing invalid actions prior to execution. For critical decisions, a human element confirms the context to ensure valid outcomes. Furthermore, actions only proceed if they pass through a structured verification process, while verified facts are logged systematically, discarding unverifiable information.
This shift offers crucial advantages in travel-related services. Proposals now have to navigate through a verification process before being executed. Unlike traditional systems where the booking and its confirmation are performed by the same agent, SCL allows for proposals to require clearance before action is taken. Consequently, travel agents have the capability to recommend changes like rebooking due to sudden schedule alterations but must obtain user consent before processing those changes.
The structured approach also ensures that all decisions are documented, distinguishing between arbitrary decisions made by AI and clear records of logic and reasoning. This transparency creates a notable advantage for travelers, who will be able to understand the rationale behind bookings made on their behalf. Additionally, this method aligns with emerging regulatory requirements, such as those outlined in the EU AI Act, ensuring that both businesses and travelers will benefit from an auditable system.