Gateway Q&A Service

Natural-language Q&A over gateway MongoDB data using Claude AI.

How it works

1. On startup, every .js file in models/ is loaded and its Mongoose schema is parsed into a plain-text summary.
2. POST /ask accepts a plain-English question.
3. Claude call 1 (Query Generator) receives the schema summary and the question, and returns a raw JSON { collection, query } — either an aggregation pipeline or a find filter.
4. The query is executed against MongoDB using the native driver (read-only).
5. Claude call 2 (Answer Formatter) converts the raw results into a concise 2-3 sentence human-readable answer.

Setup

npm install
cp .env.example .env   # or edit .env directly

Edit .env:

MONGO_URI=mongodb://localhost:27017/gateway
ANTHROPIC_API_KEY=sk-ant-...
PORT=3000

Adding models

Drop any Mongoose model file into models/. The server reads all .js files at startup and extracts collection names and field types automatically — no hardcoding required.

Start

npm start          # production
npm run dev        # node --watch (auto-restarts on file changes)

API

POST /ask

// Request
{ "question": "Which intent is failing the most this week?" }

// Response
{ "answer": "The most failing intent this week is 'placeOrder' with 1,243 failures." }

GET /health

{ "status": "ok" }

Error responses

Situation	HTTP status
Missing/empty question	400
Claude returns invalid JSON	400 (with raw_response)
MongoDB query fails	500
Claude formatter fails	500