JavAI Workflow

Define a stateful streaming workflow in Java

Create a graph of document-processing, retrieval, and answer-generation nodes, with a conditional path and streaming response.

Before you start

• A Java project using JavAI Workflow and the APIs shown in the README.
• An AbstractStatefulBean subclass for workflow state.
• An OPENAI_API_KEY environment variable and the OpenAI streaming model builder shown in the README.
• Implementations for document transformation, vector database retrieval, answer generation, and the isEnoughData() decision.

Steps

1. Define a state bean, such as MyStatefulBean, extending AbstractStatefulBean and holding the workflow data.

2. Build an OpenAiStreamingChatModel using System.getenv("OPENAI_API_KEY") and GPT_4_O_MINI.

3. Create a document parser node, a vector database retrieval node, and a streaming answer-generation node; connect retrieval conditionally to answer generation when enough data is available, or back to document parsing otherwise.

4. Construct DefaultJAiWorkflow with the state bean, transitions, retrieval as the start node, and streaming mode enabled; call answerStream(question) to obtain a Flux response.

Guide limitations

• The README illustrates the workflow structure but does not provide implementations for the document, vector database, or answer-generation functions.
• The example does not document project setup, dependency coordinates, or how to execute the Java application.
• The streaming model example uses OpenAI; the README's broader model-agnostic statement does not specify other integrations.