Scanvice

Scanvice is a landing site and an AI-powered receipt and invoice scanner. You upload a photo of a receipt, invoice or bill, and the app hands it to Anthropic's Claude vision model to read it and return clean, structured JSON — vendor, transaction details, totals and an amount breakdown, itemised line items, a category guess and a per-field confidence score — which is then rendered in the UI and can be exported to XLSX.

The extraction endpoint#

The heart of it is POST /api/extract-receipt, which runs on the Node.js runtime. It sends the uploaded image to Claude (claude-sonnet-4-5) via the Messages API together with a strict extraction prompt, and the model returns JSON only. That JSON is parsed into a typed ExtractedReceipt schema before it ever reaches the client, so the frontend always renders against a known shape. Low-quality images still extract — they just come back with lowered confidence scores and warnings rather than failing outright.

Guardrails around the model#

Because the endpoint accepts arbitrary user uploads and calls a paid API, the server side is defensive: per-IP rate limiting, image MIME-type and size validation (10 MB max), a request timeout, and metadata-only logging — the receipt content itself is never retained. Without an ANTHROPIC_API_KEY the endpoint returns a clear server_misconfigured error instead of failing silently.

Around the scanner#

The results view renders a vendor card, a line-items table, an amount breakdown and confidence indicators, with one-click export to XLSX via SheetJS. The rest is a proper marketing site — hero, problem/solution, trust and CTA sections plus an MDX-style blog — with SEO built in through sitemap.ts, robots.ts and a dynamic OG icon. It's Next.js 16 (App Router), React 19, Tailwind CSS v4 and shadcn/ui on Radix primitives.