About this demo

Dispatch Inbox turns messy load emails and rate confirmations into clean, reviewable load records. It's a weekend MVP showing how an AI step fits into a real workflow. For the full story, read the case study.

How it works

• Extract. Paste an email or drop in a PDF. claude-sonnet-5-5 reads it (PDFs natively, text and layout) and returns load details as JSON that must match a strict schema. Missing values come back empty, never guessed.
• Review. Every field sits beside the source document and stays editable. Missing fields are flagged red, uncertain ones amber, and nothing counts until a person approves it.
• Follow up. If details are missing, a smaller, cheaper model (claude-haiku-5-5) drafts a short email asking for just those items. You copy it; the app never sends email.

What it costs

Live averages from every extraction this demo has run, using published per-token prices.

Per call: extraction 0.81¢ on claude-sonnet-5-5, follow-up draft 0.01¢ on claude-haiku-5-5.

Built with

Next.js, TypeScript, Tailwind and shadcn/ui · Supabase (Postgres with row-level security, magic-link auth, file storage) · Anthropic API with structured outputs · Vercel.

Case study · Source code on GitHub · Built by Coredelle Digital