Inspiration
For an independent workshop host, “I’m interested” is not the same as “I can attend and agree to pay.”
A pottery workshop might need four participants, but people reply with different conditions: “Saturday morning only” or “Sunday is better, but Saturday works if materials are included.” A poll can collect dates, and a checkout can collect payment. The difficult part is turning conditional interest into one specific plan that people actually accept.
GatherReady explores that connection. Existing booking systems already support minimum-participant payments; our focus is helping participants clarify their conditions before checkout, then keeping the accepted plan and payment commitment aligned.
What it does
GatherReady helps small workshop hosts coordinate a viable group without treating tentative interest as consent.
Our demonstration follows a pottery workshop priced at $45 per person, with three candidate times, a minimum of four participants and a capacity of six.
- Participants describe their availability in natural language or use a structured form.
- AI turns a sentence into an editable summary of requirements, preferences and ambiguities. The participant reviews and confirms it.
- Deterministic scheduling code evaluates confirmed preferences against the available times. Interest, plan compatibility and payment authorization remain separate states.
- When a condition is ambiguous, the application can ask a focused clarification based on the workshop information.
- The organizer prepares a specific offer with time, timezone, price, inclusions and cancellation terms for participants to review. The payment workflow is designed around PayPal authorization and a separate collection step once the group qualifies.
Current verification status: Live AI extraction has been demonstrated with a local model. A genuine PayPal Sandbox UI test completed buyer approval, return to the app, authorization and a USD 45 capture using pre-seeded participants and an offer. The submitted video separates clearly labelled simulated-payment UI footage from independent real Sandbox API evidence. The new workshop/invitation entry was tested separately; we do not claim one fully verified from-zero invitation-to-payment journey. Public deployment, public webhook delivery, void and refund recovery remain unverified. No real money is involved.
How we built it
We built the interface with React 19, Vite and TypeScript, and the server with Fastify, SQLite and Zod.
The live AI demonstration uses Qwen2.5:7b-instruct through Ollama. Model output is validated as structured data and presented for human review. The model does not determine payment amounts, change accepted terms or initiate payments.
Scheduling, eligibility and financial state transitions are handled by deterministic application code. The implementation includes a PayPal Orders and Payments API adapter alongside a separate simulation adapter. Specific offer versions preserve the terms participants are asked to accept.
Challenges we ran into
The central challenge was preserving the difference between understanding a request and obtaining permission. An AI interpretation can be wrong, so a suggested change must not replace a participant’s confirmed preference without explicit confirmation.
We also needed to distinguish people who are interested, people whose conditions fit a plan, and people who have authorized payment. These are different levels of commitment.
Payments across multiple buyers are not atomic. The product therefore needs explicit failure and recovery paths, including releasing unused authorizations and handling completed captures. Those paths still require real Sandbox verification.
Accomplishments that we're proud of
Our strongest verified milestone is a live-model preference revision. A participant entered: “I can do Saturday morning only, starting at 10.” The model produced a reviewable interpretation in approximately 1.5 seconds. After explicit confirmation, the deterministic planner updated the matching group from three participants to four.
The previously saved preference remained intact during preview. Cancelling a later edit and refreshing preserved the confirmed version.
This demonstrates one real interaction; it is not a general accuracy benchmark or evidence of user demand.
What we learned
AI is most useful here when it helps people express a condition, not when it makes a commitment for them.
We also learned that passing implementation tests does not establish market demand. Whether the product reduces organizer workload or improves conversion needs testing with actual workshop hosts and participants.
What's next for GatherReady
Our next priority is validating the complete from-zero invitation-to-payment journey, public webhook delivery and multi-buyer cancellation/refund recovery. Authorization and capture have already been validated in a separate genuine Sandbox UI test.
Next, we plan to test the experience with small workshop hosts and compare natural-language entry with a structured form. We want to measure correction effort, clarity of consent and whether organizers can distinguish a plausible group from a funded booking.
Built With
- fastify
- ollama
- paypal
- react
- sqlite
- typescript
- vite
- zod
