Build an Education‑Focused RAG Solution Using Alibaba PAI

Retrieval‑Augmented Generation (RAG) for Education with Alibaba PAI

RAG combines retrieval and generative AI to provide accurate, context‑relevant answers, especially in domains like law. This guide shows how to build a large‑model RAG solution for education using Alibaba PAI.

1. Prepare the Knowledge Base with PAI‑Designer

Create a knowledge base in PAI‑Designer. Prepare data (e.g., PDFs of legal texts) that meet the platform’s format requirements and upload to an OSS bucket.

Example data can be downloaded via

wget https://atp-modelzoo-sh.oss-cn-shanghai.aliyuncs.com/release/solutions/rag/data/%E6%B3%95%E5%BE%8B%E6%96%B0%E9%97%BBpdf.zip

2. Deploy LLM and Embedding Models

In the PAI console, go to Model Gallery, select a large‑language model (e.g., Tongyi Qianwen 2.5‑7B‑Instruct) and a Chinese embedding model (bge‑large‑zh‑v1.5), and deploy them. Record the VPC address and token for each service.

3. Create LLM and Embedding Connections in LangStudio

In LangStudio, create a new application flow, add connections for the deployed LLM and embedding services using the recorded base_url and api_key.

4. Build the RAG Workflow

The workflow consists of the following nodes:

rewrite_question – rewrites the user query.

retrieve – calls the Milvus vector store to fetch relevant documents.

threshold_filter – filters retrieved documents by similarity.

generate_answer – generates the final answer using the LLM.

Configure each node with the appropriate connections and parameters (e.g., chunk size, overlap, similarity metric).

5. Deploy the Online Application

After configuring the workflow, start the runtime, choose a machine type, and deploy the RAG application. Users can then interact with the model through the built‑in chat interface.

6. Case Comparison

Two legal‑domain examples demonstrate the advantage of RAG: answers generated with a plain LLM contain vague or inaccurate statements, while the RAG‑enhanced answers cite specific legal provisions and provide precise penalties.

By following these steps, you can construct a complete RAG solution on Alibaba PAI for education or other specialized scenarios.