Build a Powerful AI Search RAG Application with PAI‑LangStudio, Qwen3 & Elasticsearch

Practice Background

The article explains how to use Alibaba Cloud's AI platform PAI, the large‑model development environment PAI‑LangStudio, and the Qwen3 inference model to build an AI Search RAG (Retrieval‑Augmented Generation) application that combines Elasticsearch full‑text and vector search.

Prerequisites

Create a dedicated VPC, subnet and security group (see linked guides).

Log in to the PAI console, create or select a workspace.

Obtain endpoint and API‑Key from the AI Search Open Platform.

Log in to the Elasticsearch console, create an instance and enable HTTPS for secure access.

Deployment Steps

Step 1: Add Model Service Connection in PAI‑LangStudio Navigate PAI → PAI‑LangStudio → Connection → Model Service → New Connection, choose the AI Search Open Platform Embedding model service, and fill in the endpoint and API‑Key.

Step 2: Create Database Connection In PAI‑LangStudio select Connection → Database → New Connection, enter the Elasticsearch instance address, username and password (use http:// if HTTPS is disabled), and confirm.

Step 3: Configure Runtime Open the Runtime tab, click New Runtime, set the working path (OSS bucket directory), and attach the same VPC, subnet and security group used by Elasticsearch.

Step 4: Create Knowledge Base In the Knowledge Base tab, create a new knowledge base, specify the OSS source path for documents and an OSS output path for parsed chunks and index data, select the Embedding model service created in Step 1, the vector database connection from Step 2, and the runtime from Step 3.

Step 5: Use the Knowledge Base in an Application Flow Create a new Application Flow from the RAG template, add a Knowledge Base Retrieval node (select the index and set filters), add a Large Model node (choose Qwen3, configure parameters, enable/disable thinking mode), and run the flow with a query to see retrieval results.

Step 6: Deploy the Model Service as an EAS Instance From the Application Flow, click Deploy, choose an appropriate instance type and the same VPC, then confirm to create a PAI‑EAS model service that can be called via API.

Solution Value

By integrating the AI Search Open Platform’s embedding and rerank capabilities with Elasticsearch’s hybrid full‑text and vector search, and leveraging Qwen3’s advanced reasoning and agent abilities, developers can quickly build a one‑stop AI Search RAG application that delivers higher accuracy, up‑to‑date knowledge, and robust enterprise‑grade performance.