Skip to content

DEMO_SCRIPT.md

Latest commit

 

History

History
33 lines (24 loc) · 1.65 KB

DEMO_SCRIPT.md

File metadata and controls

33 lines (24 loc) · 1.65 KB
Raw

Demo Script

Target length: 60 to 90 seconds.

Opening

This is an AI receipt-to-form auto-fill app for the AI Intern assessment. It uses Gemini Vision to extract structured receipt data, then keeps a human review step before any result can be downloaded.

Walkthrough

  1. Open the app and point out the receipt extraction queue.
  2. Upload two receipt images to show batch handling.
  3. Select the files and click Extract.
  4. Show item-level progress and status changes.
  5. Open a receipt preview to confirm the uploaded image.
  6. Expand a complete result and show merchant, receipt type, currency, total, date, confidence, notes, and line items.
  7. Show the download icon appears only for complete results.
  8. Show a Needs review result if available.
  9. Expand it and use the inline receipt preview beside the form to manually fix missing fields.
  10. Save changes and show the row becomes Edited with a download action.
  11. Download selected complete JSON results.

Technical Talking Points

  • The Gemini API key is protected behind a server-side Next.js route.
  • The app validates file type, file size, model output, and required fields.
  • Low-confidence or incomplete extraction does not become downloadable until reviewed.
  • Default automated tests mock Gemini, so CI is deterministic and does not depend on quota or model uptime.
  • A separate live Gemini smoke test validates the real integration before final demo recording.

Closing

The goal is not only to call an AI API. The goal is to make an AI-assisted workflow reliable: validate inputs, normalize model output, handle uncertainty, protect secrets, support manual correction, and test the important paths.