Tathyanka

Problem

Business stakeholders needed insights from PostgreSQL databases but lacked SQL knowledge. Every question became a ticket for the data team, and the queue was always two days deep.

Approach

An LLM-powered NL-to-SQL pipeline with schema context injection:

1. Schema retrieval — only the tables relevant to the question are pulled into context
2. Query generation — the model emits SQL, which is parsed and validated before execution
3. Guardrails — read-only role, statement timeout, and a row-count ceiling
4. Rendering — results become Chart.js visualisations in a React dashboard

-- generated from: "revenue by region last quarter"
SELECT region, SUM(amount) AS revenue
FROM orders
WHERE created_at >= date_trunc('quarter', now()) - interval '1 quarter'
GROUP BY region ORDER BY revenue DESC;

Outcome

In production at DalloTech. Reduced data request turnaround from 2 days to under 5 minutes, with average response time below 2 seconds across 5+ data connectors.