Parenting QA

AI Q&A web app for Indonesian parents grounded in official pediatric resources (Buku KIA 2024, IDAI recommendations, Permenkes guidelines).

Features

• Grounded Q&A — single-turn answers with inline citations to source documents
• Emergency Gate — deterministic red-flag symptom detection (Tanda Bahaya) that bypasses AI and shows an emergency banner
• Scope Filtering — cosine similarity cutoff (≥0.65) to reject out-of-domain queries
• Ingestion Pipeline — CLI script to parse official PDFs, chunk, embed, and store in pgvector

Stack

• Framework: TanStack Start (React 19 + Nitro server)
• Database: PostgreSQL + pgvector
• AI: Gemini (gemini-embedding-001) via TanStack AI / Google GenAI SDK
• Styling: Tailwind CSS v4
• ORM: Drizzle
• Package manager: pnpm

Setup

Prerequisites

• Node.js 22+
• pnpm 10+
• Docker (for PostgreSQL + pgvector)

Environment

cp .env.example .env

Fill in:

Variable	Description
POSTGRES_USER	Database user
POSTGRES_PASSWORD	Database password
POSTGRES_DB	Database name
GEMINI_API_KEY	Google Gemini API key

Database

docker compose up -d        # start postgres + pgvector
pnpm db:migrate             # run migrations

Ingest documents

Place official PDFs in indonesian_parenting_resources/, then:

pnpm ingest

Development

pnpm dev        # start dev server

Production

pnpm build
pnpm start

Or via Docker:

docker compose -f docker-compose.yml -f docker-compose.prod.yml up -d

Commands

Script	Description
pnpm dev	Start Vite dev server
pnpm build	Production build
pnpm start	Start Nitro server
pnpm test	Run Vitest tests
pnpm typecheck	TypeScript check
pnpm lint	Biome lint
pnpm format	Biome format
pnpm check	Biome check (lint + format)
pnpm check:fix	Biome auto-fix
pnpm db:generate	Generate Drizzle migrations
pnpm db:migrate	Run Drizzle migrations
pnpm db:push	Push schema to database
pnpm ingest	Run PDF ingestion pipeline

Architecture

User Query
  → SafeParentingAgent (Tanda Bahaya check)
    → KnowledgeBase (pgvector similarity search, cutoff ≥0.65)
      → ParentingAgent (LLM generation with citations)
        → Response

For scope-miss queries (similarity < 0.65) or emergency matches, the LLM is never called.

Out of Scope (v1)

• Growth calculator (z-score classification)
• Multi-turn conversations
• User accounts / child profiles
• Postpartum mental health