Inspiration
Dental procurement often starts as an informal message: a clinic needs several supplies, but the request still has to become specific products, quantities and a total the buyer understands. Dental Listo AI explores a practical assistant for that step.
What it does
A buyer describes a need in English or Spanish. The assistant proposes products from a small synthetic dental catalog, asks for clarification when a request is ambiguous, and presents an editable cart. The server calculates prices and shipping. The buyer reviews and explicitly approves the exact quote before the PayPal Sandbox checkout can proceed.
For example, two restoration kits and one prophylaxis kit total USD 120.00, including USD 5.00 shipping. Changing the cart invalidates the previous approval.
How we built it
The isolated prototype uses Next.js, React, TypeScript, Node.js and SQLite. An optional OpenAI GPT-5-mini adapter uses structured extraction through the Vercel AI SDK and Zod validation. Model output expresses intent; the application's deterministic catalog and checkout logic control products, prices, approvals and payment state.
The PayPal integration uses Sandbox OAuth and Orders API v2, buyer approval, stable idempotency keys, session-bound order history, capture verification and read-only reconciliation. Signed-webhook validation and a webhook-only relay are implemented. PayPal credentials remain server-side.
Codex helped implement and review the application, regression tests, reproducible verification scripts, English demo materials and the standalone source export.
What makes it useful
- A reviewable cart grounded in a known catalog.
- Explicit clarification instead of invented products.
- Human approval tied to the current quote.
- Persistent payment history and controls against repeated capture attempts.
- A clearly labeled local simulation for exploring the interface without credentials.
The prototype uses four fictional kits and synthetic prices. It is separate from the commercial Dental Listo application and contains no customer data.
Verification and current limitations
Five genuine OpenAI extraction cases were verified on 17 October 2026 in an earlier adapter revision. Subsequent guard changes have offline test coverage. The existing 82-second English preview shows SIMULATED interpretation.
PayPal Sandbox OAuth and order creation have historical evidence. A buyer-approved completed capture and authentic signed-webhook delivery have not yet been verified. Historical orders are not presented as usable current checkout links. The complete AI-to-payment journey remains pending.
There are no measured customer-adoption or time-saving claims. Public source, the final public YouTube video and the final competition declarations are still being prepared. This text is a project draft.
Testing instructions
Use Node.js 22.13 or later. From the isolated application directory, run npm ci, then npm run verify:local. This verification disables external AI and PayPal payments. To explore the UI, run npm run dev and open http://localhost:3100.
Enter “I need two restoration kits and one prophylaxis kit,” review the proposed products, add them to the cart and approve the quote. Edit a quantity to check that approval is invalidated. Ambiguous or unavailable products should require clarification. Live PayPal Sandbox testing requires separate test credentials and the buyer's own approval.
Prior work and hackathon contribution
Dental Listo existed before the hackathon as a separate commercial procurement project. During the hackathon we built this isolated AI-and-PayPal laboratory: structured intent extraction, a synthetic catalog, human approval, persistent Sandbox checkout controls, automated tests, an English interface and demo materials, and a sanitized standalone export. The commercial application, deployment and data are outside this entry.
Built With
- next.js
- node.js
- openai
- paypal
- react
- sqlite
- typescript
- vercel-ai-sdk
- zod