Turning Mess into a Fixed Shape

the extraction job

Part of: Structured JSON Extractor

Raw text is everywhere: a signature block at the bottom of an email, a receipt someone pasted into a chat, a paragraph of meeting notes. A human reads it fine. Software can't do anything with it until it has structure. A structured extractor closes that gap. It reads messy free text and returns the same fixed set of fields every time, as JSON your program can trust. What we're building Over eight lessons you'll build a tool that turns messy text into validated JSON: contacts (name, email, phone) and invoices (id, total, line items). The output is never a paragraph. It's a strict object with known keys, ready to drop into a database row or a spreadsheet cell. The one idea: a target schema Everything starts with the schema , the exact shape you demand back. For a contact: The schema is a contract. It names every field, so the model knows what to look for and your code knows what to expect. Without one, the model returns a slightly different shape on every call and your parser is left guessing. The loop, with the work in two steps It's the same six-step loop from the playbook. The interesting work happens in prompt and parse : 1. Input : the messy text. 2. Prompt : "extract exactly th

Challenge: Required Field Audit