Bright Path Tuition (AlloyDB + MCP)

An AlloyDB-backed AI assistant for a tuition centre. The owner asks "who hasn't paid this month" in plain language and gets the answer back in a paragraph.

Small businesses tend to have data — Excel exports, POS dumps, attendance apps — but no time or skill to query it. So the data sits unused. The owner is a teacher, not a database engineer.

The architecture stores the data in AlloyDB, exposes typed query tools via MCP (Model Context Protocol), and lets an LLM translate natural-language questions into the right tool calls.

• AlloyDB Postgres holds students, attendance, payments, exam results
• An MCP server exposes typed tools — get_unpaid_students, attendance_below_threshold, top_performers_by_subject
• The LLM picks the right tool, fills in parameters, executes, and explains the answer in plain language

The discipline that MCP imposes is the part that makes this safe. The LLM doesn't write SQL. It picks from a fixed menu of typed queries. No hallucinated joins, no leaked rows, no surprise full-table scans. Determinism by design.

A representative ask: "Show me students who missed last week's class and haven't paid this month." Two MCP tool calls, one paragraph back, no data team in the loop.

For a teacher running a tuition centre, this is the only realistic path from "I have the data" to "I can ask it questions" without hiring an analyst.