Solution Design

Solution Design

In academic research scenarios, users need to pore over an article for further research. They use foundation models to delve into the material.

After uploading a document on the AI research assistant frontend page, the user asks questions to the LLM based on the document content. This triggers the preset workflow in the agent and enables interaction with the LLM.

Workflow Node Design

Key nodes in a workflow should be selected. Each node is responsible for a specific task. The functions and design principles of each node are detailed as follows:

• Start node: The Start node serves as the entry point for a workflow, receiving text input from users.
• Branch node: A Branch node functions like an IF-ELSE statement, enabling multi-branch conditional judgments. It is used to design branching processes and implement logic decision-making. For example, it can determine whether a user has uploaded a document or whether any user has uploaded a document.
• Plugin node: A Plugin node is the core component for calling third-party capabilities in a workflow. As an important carrier for functional extension, it enables the execution of specific tasks by calling designated plug-ins. Each plug-in is essentially a set of API tools encapsulated in a standardized format. Plug-ins provide plug-and-play modular services, extending the capabilities of workflows to handle more complex tasks. For example, the pre-installed Read_File plug-in can read documents uploaded by users.
• Knowledge Repo node:
  • The Knowledge Repo node allows users to configure knowledge information.
  • It retrieves relevant knowledge based on user-entered questions.

• LLM node: The LLM node allows users to use the capability of LLMs to configure a deployed model, write prompts, and set parameters to perform specific tasks. For example, the LLM node can extract and output document content, generate and refine answers, and format the output accordingly.
• Aggregation node: The Aggregation node can aggregate the outputs from multiple branches, facilitating unified configuration of subsequent nodes.

  In a workflow with two branches, an Aggregation node is required to consolidate the outputs of both branches. In this scenario, you can use the Aggregation node to aggregate the output variables from multiple branches. The Aggregation node reads the first non-null value from these branches, which is then passed to downstream nodes for further processing. You do not need to handle the outputs of branches that are not running, simplifying data flow management.

• End node: The End node is the final node of a workflow and outputs the final result.