Generative AI powered by your SQL data

Video script:

– Go beyond keyword search or T-SQL similarity search, whether on-premises or in the cloud, by combining SQL data with Azure AI Search and Azure OpenAI services on Azure to power intelligent natural language search for your SQL workloads, and use generative AI to modernize experiences with personalized, conversational-tone, and accurate responses to natural language queries in e-commerce engines, financial apps, CRM systems, and more. This leverages an approach called Retrieval Augmented Generation, which allows you to define SQL data as a trusted source.

– Then, when a user submits a prompt, you can use it to retrieve and augment additional information, providing a large language model with additional context that can infer and generate more relevant and accurate responses. This process is called grounding. All data you present to the language model is on the side and is not absorbed by the model. It also inherits the fine-grained access management controls of SQL Server. Also, depending on who is querying the data, you can retrieve information and generate and serve responses based on the user’s data access permissions.

– But you need to go beyond keyword search or T-SQL-like search while searching. Enabling natural language search using vectors is the key to making everything work. This is possible by using Azure AI Search as a managed search index with Azure OpenAI service to build and maintain the index by generating vector embeddings based on defined fields in your database.

– User prompts are also vectorized in real time, and the query engine uses GPS-like coordinates to find the most similar parts between what is queried in the prompt and the data in the database. It is easy to connect vectorized SQL data to search on the ground-generated AI responses.

-Use Azure OpenAI Studio with a large-scale language model deployed on Generate AI here. Select Get Data, select Azure AI Search as the data source, and select Vector Search on Index.

– You can also choose search options, and hybrids allow you to combine keyword and vector searches. There you can test the generated AI responses. You can generate AI responses that work with your data in the Azure OpenAI Studio Playground, and even instruct your AI assistant on how to act on system messages.

– Then try out sample prompts to test whether the generated responses are relevant and accurate based on vectorized data in your index. This is how the combined power of SQL data and AI in Azure will bring SQL workloads into the GenAI era.