SQL

SQL is a classic example of a declarative language, where users declare what data they want and the database query optimizer decides the optimal execution plan on its own.

AI agents typically generate database queries or API calls dynamically via large language models, rather than using parameterized queries pre-written by engineers.

The SQL language is divided by function into four major categories: DDL, DML, DCL, and DQL.

Merely practicing SQL and probability interview questions is insufficient for transitioning from academia to industry data science roles

Common window functions include ROW_NUMBER(), RANK(), LAG()/LEAD(), and SUM() OVER().

SQL learning can be divided into three stages: beginner covers single-table queries and basic aggregation; intermediate covers multi-table joins, subqueries, and CASE WHEN; advanced includes window functions, CTEs, recursive queries, and performance optimization.

Using the wrong JOIN key in a one-to-many relationship causes row multiplication, making SUM aggregation results grow by multiples, while SQL does not throw any exception

Having LLMs directly generate SQL is high-risk, because the model does not understand table structures, field meanings, or business definitions, and easily generates queries that are syntactically correct but produce wrong results.

TypeScript、Java等常规编程语言的AST语法树通常在一两千行左右，SQL语言的语法树大概在一千行左右

Semantic parsing refers to converting natural language statements into machine-executable logical forms, such as SQL queries, SPARQL graph queries, or program code.

Much semantics cannot be derived from syntax and must be measured empirically; the model initially did no optimization at all in query optimization and only improved once performance benchmarks and real datasets were provided.

dbchat lets users ask questions in natural language, and the AI agent generates SQL in real time and returns the results.

The declarative pipeline pattern traces back to the tradition of SQL and Makefiles: declare targets and dependencies and let the executor decide the steps.

Domain-specific languages (DSLs) are programming languages designed for a specific problem domain, such as SQL for database queries, regular expressions for text matching, and Verilog for hardware description.

croft can pull data from APIs and files, transform it with SQL or TypeScript, and refresh data on a schedule.

When limiting row counts, the entire query should be wrapped as a subquery and an outer LIMIT applied, rather than appending LIMIT, so that the outer limit always takes effect and the model cannot break through via a large inner LIMIT.

Evaluating the correctness of LLM-generated SQL should compare result sets rather than SQL text, rounding floats to 6 digits before comparison to avoid tiny discrepancies.

A cost-proxy method for detecting Cartesian products: count the row count of each referenced table and sum them, switch to multiplying when a join without ON or USING is encountered, and reject the query if the threshold is exceeded.

Simple string checks (such as sql.strip().lower().startswith('select')) fail in at least three cases: semicolon-chained statements, data-modifying statements hidden in CTEs, and reading files via table functions in DuckDB.

Relational SQL performance drops sharply for join queries of more than 5 hops, and multi-table nested JOINs see latency spike under concurrent emergency queries.