AI Data Analyst: Ask Your Database Anything in Plain English
An AI-powered data analyst that turns plain English questions into SQL queries, charts, and board-ready PDF reports — with full reasoning transparency. Try the live demo.
Try it yourself — no signup required
Pre-loaded with Nexora, a realistic B2B SaaS dataset: 480 customers, $110K MRR, 10 tables, 15 countries.
The Problem
Every SaaS company has the same bottleneck: data locked in databases that only engineers can query. The CEO asks "what's our net revenue retention?" and waits for someone to write SQL. The head of sales wants pipeline conversion rates by stage — that's a ticket in the backlog.
Business teams shouldn't need SQL skills to answer their own questions. And engineering teams shouldn't spend hours writing one-off queries that get used once.
What We Built
An AI data analyst that anyone on the team can use. Type a question in plain English — the AI figures out what data you need, writes the SQL, runs it against your database, and explains the results. No SQL knowledge. No waiting for engineering.
- Natural language → SQL → Results — ask "which customer cohort has the best retention?" and get the answer with the SQL shown for verification
- Auto-generated charts — revenue trends, cohort curves, funnel breakdowns, and regional comparisons rendered automatically
- Board-ready PDF reports — MRR, ARR, churn rates, pipeline health, and regional breakdowns compiled into a downloadable PDF
- Churn & risk analysis — correlates usage drops, support tickets, and health scores to flag at-risk accounts
- Upload your own data — drag and drop a CSV or SQLite file and start querying in seconds
Why reasoning transparency matters: Most AI analytics tools are black boxes — you get a number but can't verify it. That's a dealbreaker for anyone making decisions on data. Our tool shows every step: how it interpreted your question, what SQL it wrote, why it chose that approach. You can verify the logic before acting on the answer.
How It Works Under the Hood
The system uses Claude as an AI agent with specialized tools exposed via MCP (Model Context Protocol). When a user asks a question, the agent orchestrates a pipeline:
- Schema discovery — the agent inspects the database structure and documentation to understand what data is available
- SQL generation — writes a query tailored to the user's question and the actual schema
- Execution & validation — runs the query in a sandboxed environment (read-only, no data modification)
- Analysis — interprets the results, generates charts if useful, and explains what the numbers mean in business context
For complex requests like "generate a board report," the agent chains multiple queries and generates Python visualizations and PDF output — all autonomously.
Tech Stack
- Claude Agent SDK — agent orchestration with real-time streaming responses
- MCP (Model Context Protocol) — 7 tools: schema discovery, SQL writing, SQL execution, Python scripting, chart generation, and PDF reports
- Next.js 16 — full-stack framework with SSE streaming and React frontend
- better-sqlite3 — embedded database, zero infrastructure setup
- Cloud Run — containerized deployment with auto-scaling
Sample Queries You Can Try
The demo is pre-loaded with Nexora — a realistic B2B SaaS company. Here are questions that showcase different capabilities:
What's our MRR trend over the last 12 months? Break it down by new, expansion, and churned revenue.
Which customers are most likely to churn? Show accounts with declining usage and low health scores.
Compare customer acquisition cost and conversion rates across all marketing channels.
Generate a board-ready PDF with MRR, ARR, net revenue retention, top 10 accounts, and revenue by region.
The Result
A question that used to require an engineer writing SQL and a 24-hour turnaround now takes under 30 seconds. Business teams can self-serve their own data questions. Engineering stops getting one-off query requests.
The tool is config-driven — we can deploy a customized version for any team by connecting their database and editing two JSON config files. Same architecture, different data, different branding. The same approach powered the AI-Powered Trading Analytics project we shipped for a fintech client.
See it in action
The demo is live — pre-loaded with a realistic SaaS dataset. No signup, no credit card. Just ask a question.
Want something like this built?
Tell us the problem. We'll tell you what 72 hours can produce.