Phase 1 Invoice Extractor

This is a Next.js 16 app for a focused v1 invoice extraction workflow:

• Upload a single PDF, PNG, or JPG invoice
• For PDFs, render up to the first 5 pages as images
• Send the document to Gemini through the Vercel AI SDK
• Validate the result against a strict Zod schema
• Review the extracted fields, line items, and raw JSON in the UI

Environment

Create a local env file and add your Google AI Studio key:

cp .env.example .env.local

GOOGLE_GENERATIVE_AI_API_KEY=your_google_ai_studio_api_key
GEMINI_MODEL=gemini-3-flash-preview

GEMINI_MODEL is optional. The app defaults to gemini-3-flash-preview.

Run locally

pnpm dev

Open http://localhost:3000.

What this v1 does

• Invoice-only extraction
• PDF/image upload through a Next.js route handler
• Page-image-heavy PDF processing with a PDF-file fallback
• Strict JSON output using generateText() plus Output.object()
• shadcn/ui upload and results interface

What this v1 does not do yet

• Persistent file storage
• Background job queues
• Contract extraction
• Cross-document search or embeddings