// Field Notes

Blog

Practical notes from client work: data engineering, custom software, and AI enablement for small and mid-sized businesses. What we build, what breaks, and the numbers behind it. No hype, no vendor slides.

Part 4 / 4

Text to SQL That Actually Works

Text to SQL with LLMs, Part 4: Production-Grade Systems: Semantic Layers, Agents, Guardrails, and Evals

How production text to SQL systems work: semantic layers, agentic self-correction, read-only guardrails, MCP integration, and the eval harness that proves accuracy.

July 23, 2026 · 10 min read

Part 3 / 4

Text to SQL That Actually Works

Text to SQL with LLMs, Part 3: Retrieval, Golden Queries, and Business Context

Static schema context has a ceiling. Learn how retrieval, verified question-SQL pairs, a business glossary, and value grounding push text to SQL accuracy higher.

July 22, 2026 · 9 min read

Part 2 / 4

Text to SQL That Actually Works

Text to SQL with LLMs, Part 2: Schema Context Engineering That Actually Moves Accuracy

The biggest accuracy lever in text to SQL is how you present your schema. Column descriptions, sample values, relationships, and curated views, with SQL to build them.

July 21, 2026 · 10 min read

Part 1 / 4

Text to SQL That Actually Works

Text to SQL with LLMs, Part 1: Why the Baseline Approach Fails on Real Databases

Basic schema-in-prompt text to SQL still fails on real databases in 2026. See the current benchmark numbers, the five failure modes, and what actually fixes them.

July 20, 2026 · 11 min read