AI Refinery™ 101¶

What is AI Refinery™¶

AI Refinery™ by Accenture is a Cloud Service for developing and executing AI multi-agent solutions. It is a platform designed to help organizations:

• Adopt and customize large language models (LLMs) to meet specific business needs.
• Integrate generative AI across various enterprise functions using a robust AI stack.
• Foster continuous innovation with minimal human intervention.

This platform ensures seamless integration and ongoing advancements in AI capabilities within your organization.

Key Features of AI Refinery™¶

Agentic Framework¶

The Agentic Framework within AI Refinery™ offers a structured approach for deploying AI agents built on top of LLMs. These agents can autonomously perform tasks, make decisions, and interact with other agents and systems. Additionally, the framework includes advanced algorithms for chain-of-thought reasoning, dynamic structured memory management, query transformation, and more.

Pre-built Utility Agents¶

Provides a suite of ready-to-deploy agents capable of performing real-time web searches, data analysis, document writing, and more. Deployment is as simple as configuring a YAML file.

Custom Utility Agents¶

In addition to the pre-built utility agents, you can effortlessly create and integrate your custom agents into the AI Refinery™ Agentic Framework with just a few lines of code.

Comprehensive Model Catalog¶

AI Refinery™ features an extensive model catalog, including LLMs, VLLMs, rerankers, and more. You can choose any model from our model catalog for both the pre-built and your custom utility agents.

Flexibility¶

Easily integrate existing agents and GenAI stacks like LangChain, LlamaIndex, and Firefly. For example, you can seamlessly add a LangChain-based RAG agent to your YAML file. Additionally, you can seamlessly update default LLMs, embedding endpoints, and more using models from our catalog.

Components of AI Refinery¶

Distiller¶

The Distiller framekwork of our SDK is designed to streamline complex workflows by orchestrating various agents that handle different tasks. Below are the key components of the Distiller and their roles. AI Refinery SDK is equipped with a distiller client api. An overview of our Distiller framework can be seen in the figure below.

Orchestrator¶

The Orchestrator is a general-purpose component that routes users' requests i.e., queries to the appropriate agents. It ensures that tasks are assigned to the correct agents based on their capabilities and the nature of the request.

Features¶

• Request Routing: Directs requests to the suitable agent based on the user query and the description of each agent.
• Task Management: Manages the lifecycle of tasks from initiation to completion.
• Scalability: Efficiently handles multiple requests and scales as needed.

Super Agent¶

A Super Agent is a general-purpose agent that decomposes a complex task into multiple subtasks. These subtasks are then assigned to the appropriate utility agents. Super Agents are configured by the user to handle specific types of complex tasks.

Features¶

• Task Decomposition: Breaks down complex tasks into manageable subtasks.
• Custom Configuration: Users can configure Super Agents based on their specific needs.
• Integration: Works seamlessly with other agents to complete tasks efficiently.

Utility Agents (Configured/Built-in)¶

Utility Agents are built-in agents that can accomplish a single task such as research, analytics, writing, etc. These agents are pre-configured but can be adjusted by the user to meet their requirements.

Features¶

• Single Task Focus: Each Utility Agent is specialized for a particular task.
• User Configuration: Allows user adjustments to better fit the project needs.
• Efficiency: Optimized to perform their specific tasks quickly and accurately.

Utility Agents (Customized)¶

Customized Utility Agents are user-defined agents that leverage proprietary data or tools within their own environment. These agents are tailored by the user to perform specific tasks using their own data sets and tools.

Features¶

• Customization: Fully customizable to use proprietary data and tools.
• Environment Specific: Operates within the user's own environment.
• Flexibility: Provides the flexibility to define unique tasks and workflows.

AI Refinery SDK¶

The AI Refinery SDK is engineered to facilitate the development of AI solutions by leveraging the aforementioned Distiller. An overview of the AI Refinery SDK workflow is illustrated in the figure below. As we can see in the figure, it is comprised of two main components which are detailed below.

Server Backend¶

The server backend of the AI Refinery SDK is responsible for hosting the Distiller, which includes the orchestrator, super agent, and built-in utility agents. Additionally, it instantiates and hosts user-provided custom utility agents. The server backend also manages and stores user account history in a dedicated database.

User Backend¶

The user backend of the AI Refinery SDK is equipped with comprehensive APIs to facilitate the creation of custom utility agents and to enable seamless communication with the server backend for queries and requests. Additionally, the user backend is responsible for managing proprietary data that users wish to employ as the knowledge base for their agents.