Powering connections and smart operations in DeltaV
The AI Engine stands at the core of DeltaV ↗️ (opens in a new tab) and its features, as it allows users and developers to connect to a wide range of agent-based services. Once an agent is registered ↗️, the offered service is visible to the AI Engine and it can start connecting users and services.
This system is equipped with personalized capabilities, supported by an internal agent that performs tasks efficiently. An internal agent is created by the AI Engine and made available for communication via the DeltaV user interface. The AI Engine interprets the human text input provided to the agent and starts working asynchronously on your behalf as soon as it receives your intent. This customized method uses Large Language Models (LLMs), which are essential for improving the AI Engine's understanding, coordination and problem-solving capabilities.
The AI Engine introduces users and developers to a unified ecosystem of agent-based services. Once an agent and its services are registered in the Agentverse ↗️ and consequently in the Almanac ↗️, they become an integral part of the AI Engine landscape and coordinate dynamic connections between users and services. For example, if you ask the DeltaV agent what the weather will be like tomorrow at your location, it will connect to a registered agent in the Almanac and receive the latitude and longitude values of your current location. The weather forecast for that day is then retrieved by the chat agent via a connection to another registered agent which specializes in weather-related services.
At the heart of the AI Engine, there is an architecture consisting of the following components:
Objectives are the encapsulation of the user's general goals, communicated in natural language via the DeltaV chat.
Tasks form a dynamic sequence of steps that drive the achievement of these defined goals. Tasks involve complicated processes, including the allocation of resources and dependencies, which are managed by agent-based services.
Subtasks are auxiliary or secondary tasks that are usually linked to and dependent on the completion of a primary task and often require a specific context or information to complete.
In this context, a task refers to an agent or service that provides a specific action or information requested by the user that is directly accessible via DeltaV. In contrast, a subtask also responds to user requests within DeltaV, but typically provides additional or complementary services that often rely on a prior context or additional information for their execution. For example, while a task might be an agent that provides the user's current account balance, a subtask might involve converting the account balance to a different currency, which requires additional context or user input.
Finding new information is a key focus of the AI Engine to significantly improve the user journey. This is crucial for the execution of services, such as booking a hotel room for your holiday in a specific city. In an environment where reservations are centralized, this seems like a simple process. However, for the booking to be successful, the AI Engine must be able to understand the user's input and objectives and communicate with multiple agents. At this stage, the AI Engine's ability to understand and plan is very important: the user's goal is broken down into a series of smaller tasks and subtasks, each representing an integral step towards the desired end result. This coordination may be automatic, or in certain situations where the AI Engine is unsure, it may require user input to confirm the task selection.
Context building plays a crucial role, allowing the AI Engine to continuously improve its understanding by transforming data. Context building is an ongoing process within the AI Engine that involves the continuous improvement of the knowledge base during the AI Engine session. In other words, context building is the continuous act of collecting and/or transforming new knowledge to complete a task.
The final step of the AI Engine is smart routing, that is the ongoing process within the AI Engine that makes it aware of all registered AI Agents and the tasks for which they are best suited for. This process takes into account the context and past performance history of these agents to guide the AI Engine's decision-making process. In this way, the AI Engine selects the most suitable agents, taking into account the agents' services and their past performance metrics. Trust becomes a key factor, favoring agents with a track record of reliable behavior. Smart routing not only ensures the completion of tasks, but also creates a sense of reliability and efficiency in the operations.