Case Study 2 - Fabrikam
Existing Environment
Azure Environment
Fabrikam has a single Azure subscription in the East US 2 Azure region. The subscription contains an Azure SQL database named DB1. DB1 contains the following tables:
* Patients
* Employees
* Procedures
* Transactions
* UsefulPrompts
* ProcedureDocuments
You store a column master key as a secret in Azure Key Vault.
You have an on-premises application named TransactionProcessing that uses a hard-coded username and password in a connection string to access DB1.
Problem Statements
Users report that after executing a long-running stored procedure named sp_UpdateProcedureForPatient, updates to the underlying data are sometimes inconsistent.
Requirements
Planned Changes
Fabrikam plans to manage all changes to Azure SQL Database objects by using source control in GitHub. Every pull request submitted to production will be validated before it can be merged.
Deployments must use the Release configuration.
Security Requirements
Fabrikam identifies the following security requirements:
* The TransactionProcessing application must use a passwordless connection to DB1.
* The Employees table contains two columns named TaxID and Salary that must be encrypted at rest.
* Auditors must have a tamper-evident history of transactions with cryptographic proof of changes to the employee data.
Database Performance Requirements
Records accessed by using sp_UpdateProcedureForPatient must NOT be changed by other transactions while the stored procedure runs.
AI Search, Embeddings, and Vector Indexing
Fabrikam identifies the following AI-related requirements:
* Queries to the ProcedureDocuments table must use Reciprocal Rank Fusion (RRF).
* Users must be able to query the data in DB1 by using prompts in Copilot in Microsoft Fabric.
* The UsefulPrompts table will store prompts that doctors can use to help diagnose patient illness by connecting to an Azure OpenAI endpoint.
Development Requirements
Fabrikam identifies the following development requirements:
* Provide the functionality to retrieve all the transactions of a given patient between two dates, showing a running total.
* Expose a Data API builder (DAB) configuration file to enable Azure services to perform the following operations over a REST API:
- Read data from the procedures table without authentication.
- Read and insert data into the Transactions table once authenticated.
- Execute the sp_UpdateProcedurePatient stored procedure.
* Provide the functionality to retrieve a list of the names of patients who underwent medical procedures during the last 30 days.
* Information for each medical procedure will be stored in a table. The table will be used with a large language model (LLM) for user querying and will have the following structure.

DAB
You create a DAB configuration file that meets the development requirements for DB1 and includes the following entities.

You implement ProcedureDocuments to support the planned changes.
When users consume data through the Retrieval Augmented Generation (RAG) pattern, they experience data retrieval delays.
You need to improve the data retrieval performance and reduce the number of tokens per retrieval.
What should you implement?

Which data type is used to store vector embeddings in SQL?

You have an Azure SQL database named SalesDB on a logical server named sales-sql01.
You have an Azure App Service web app named OrderApi that connects to SalesDB by using SQL authentication.
You enable a user-assigned managed identity named OrderApi-Id for OrderApi.
You need to configure OrderApi to connect to SalesDB by using Microsoft Entra authentication.
The managed identity must have read and write permissions to SalesDB.
Which Transact-SQL statements should you run in SalesDB?

You have a database that contains production data. The schema is stored in a Git repository as an SDK-style SQL database project and contains the following reference data.

A deployment pipeline can be rerun automatically when a transient failure occurs.
You need to deploy the reference data as part of the same CI/CD process. Rerunning the pipeline must produce the same outcome and must NOT create duplicate rows.
What should you do?

Which SQL feature is commonly used to integrate AI-generated insights?