The database is no longer just a store. It is the AI.
SQL Server 2025 ships with native vector types, built-in LLM calls, and REST endpoint invocation directly from T-SQL — collapsing the boundary between your data layer and your AI stack. For developers and architects who already know SQL, this is the most important platform shift in a decade. This workshop is your guided, hands-on introduction to all of it.
Over eight focused hours, you will move from T-SQL fundamentals through the full arc of modern AI-adjacent development — writing production-grade code at every step, not watching someone else write it. Every module ends with a lab exercise built on the same schema, so by end of day you have a working semantic search application that runs entirely inside SQL Server, no application middleware required.
You Will Build a production-ready semantic product search engine — embedded entirely within the database engine — that ingests data, generates vector embeddings via Azure OpenAI, stores them in native VECTOR(1536) columns, retrieves results using VECTOR_DISTANCE(), generates natural-language summaries with AI_GENERATE(), and logs every AI inference call to a structured audit table with retry logic, row-level security, and dynamic data masking on sensitive columns.
Foundations
• Rewriting row-by-row CURSOR logic as set-based operations and measuring the improvement on actual execution plans
• Using window functions (ROW_NUMBER, RANK, LAG, LEAD, CUME_DIST, PERCENTILE_CONT) for trend analysis and ranking without self-joins
• Choosing correctly between CTEs, temp tables, table variables, and subqueries based on row count, statistics requirements, and plan reuse
Modern T-SQL
• Parsing and producing JSON with OPENJSON, FOR JSON PATH, JSON_ARRAY(), and JSON_OBJECT()
• Implementing system-versioned temporal tables for point-in-time audit queries with zero trigger code
• Using SQL Server's native graph extensions (NODE, EDGE, MATCH, SHORTEST_PATH) for relationship traversal
SQL Server 2025 & Cross-Platform AI
• Calling any HTTPS REST endpoint — Azure OpenAI, Microsoft Foundry, Ollama local models — directly from T-SQL using sp_invoke_external_rest_endpoint
• Storing and querying high-dimensional embeddings with the native VECTOR(n) data type and VECTOR_DISTANCE() across cosine, dot product, and Euclidean metrics
• Invoking language models natively with AI_GENERATE() for row-level summarization, classification, and entity extraction
• Scoring ONNX machine learning models in-database with PREDICT() — zero data movement, no Python runtime
• Navigating capability differences between SQL Server 2025, Azure SQL Database, and Microsoft Fabric SQL Analytics Endpoint for teams running workloads across multiple platforms
Performance & Production Readiness
• Accelerating analytics workloads with clustered columnstore indexes and approximate aggregations (APPROX_COUNT_DISTINCT, APPROX_PERCENTILE_DISC)
• Designing embedding tables with DiskANN approximate nearest-neighbor indexes and pre-filter strategies that reduce distance computation by orders of magnitude
• Implementing defense-in-depth security: Transparent Data Encryption, Always Encrypted column-level cryptography, Dynamic Data Masking, and Row-Level Security in a single multi-tenant schema
• Writing production stored procedures with structured TRY/CATCH error handling, idempotent retry logic, and full AI audit trails
This session is designed for database developers, data engineers, and solution architects with working T-SQL knowledge who are ready to build AI-powered applications directly on Microsoft SQL platforms. It is a strong preparation session for the DP-800: Designing and Implementing AI-Enabled Database Solutions certification exam.