March 2026 Summaries

PyDough, a domain-specific language (DSL), addresses the gap between human reasoning and SQL by allowing users to express analytical intent in relational terms directly. It leverages a knowledge graph derived from relational databases to model the structure of data and enforce correctness in query generation. Unlike traditional natural language querying systems, PyDough does not rely on manually curated semantic layers but automatically derives its knowledge graph from the database schema, capturing relationships explicitly. This approach eliminates common errors such as invalid joins, SQL injection, and prompt injection by ensuring that only validated and structurally sound expressions can be compiled into SQL. As a result, PyDough provides a more accurate and safe execution model, offering a balance between fluency and safety in natural language analytics. The open-source project invites feedback and collaboration from teams to further develop its capabilities.