01 / PRODUCTION AI · WORK
Context-Building SQL Agent
Natural language to validated SQL, grounded in a context graph the agent builds for itself.
Instead of handing an LLM a raw schema, the agent auto-discovers the database - entities, fields, record counts, and domain-value encodings - and maintains that context graph as the single source of truth, so generated SQL never invents tables or fields. A two-tier prompt system pairs a global core engine with per-project behavioral prompts, with guardrails that keep style prompts from overriding SQL generation rules and validation.
Internal work project - walkthrough on request
customers128 fields · enriched
orders96 fields · enriched
line_items74 fields · enriching
-- "top 5 customers by revenue"SELECT c.name, SUM(o.total) AS revenue
FROM customers c JOIN orders o
ON o.customer_id = c.id
GROUP BY c.name ORDER BY 2 DESC
LIMIT 5;
FROM customers c JOIN orders o
ON o.customer_id = c.id
GROUP BY c.name ORDER BY 2 DESC
LIMIT 5;