Chat with your warehouse.
"How's revenue this quarter by channel?" "Who are the top 20 accounts at risk?" Plain-English questions, SQL under the hood, charts on the screen — governed, safe, cited.
SELECT region, SUM(mrr) FROM v_revenue
WHERE quarter IN ('Q2','Q3','2026')
GROUP BY 1How it stays safe.
In the box.
- Semantic layer (dbt/Cube)
- Column-level masking
- Cost caps per user
- Save + share queries
- Slack integration
- Weekly digest emails
- Chart library
- Exports to Sheets / CSV
Analytics AI FAQs.
Vanilla text-to-SQL is. We add a semantic layer (dbt/Cube), enforced schemas, and query validation so the model only asks questions it's allowed to.
Question taxonomy the assistant must handle
Text-to-SQL demos handle the easy 20%. Real business questions fall into five shapes with very different accuracy profiles.
| Question shape | Example | Baseline accuracy | How we lift it |
|---|---|---|---|
| Direct lookup | "Revenue last month" | 95%+ | Cached metric layer, no LLM |
| Filtered aggregate | "MRR by plan, EU only, QoQ" | 70-85% | Semantic layer with dimensions typed |
| Comparative | "Why did churn spike in July?" | 40-60% | Guided drilldown, not free SQL |
| Causal / attribution | "Which campaign drove signups?" | 30-50% | Restrict to pre-built attribution models |
| Predictive | "Will we hit Q4?" | <40% without models | Refuse or route to forecast pipeline |
The semantic layer contract
Every metric the assistant can answer needs a single source of truth. This is the minimum spec per metric.
- Canonical name + synonyms
"Net Revenue Retention" also matches "NRR", "net retention", "expansion minus churn" — no aliases lookup, no answer.
- Grain
Row-level definition (customer-month, order-line). Assistant refuses queries that violate the grain.
- Allowed filters
Explicit list of dimensions with types. LLM cannot invent columns; unknown filter returns a clarifying question.
- Owner + freshness SLA
Every metric shows who owns it and when it last refreshed — assistant surfaces this on every answer.
- Approved joins
Pre-defined join paths only; blocks cartesian products and the classic "double-counted revenue" bug.
- Redaction rules
PII columns tagged; assistant strips or aggregates before returning to unauthorised roles.
Your data. On tap. In English.
Book a demoServing businesses across Rayalaseema
Founder-led website design, development and automation — for businesses in Andhra Pradesh and beyond.
Automate what's costing your team hours
Describe the workflow — we'll reply with what's possible, what it costs, and where to start.
- Personal reply from Siddiq within 24 hours
- Honest advice — no pushy sales, no fake urgency
- Clear next steps, timeline, and estimate