$29 diagnostic playbook
WhatsApp / Airtable / OpenAI state bugs
A focused first pass for workflows where WhatsApp messages, Airtable state, OpenAI output, Telegram handoff, and reservations mostly work but break on edge cases.
First map
WhatsApp webhook -> normalize sender/session -> Airtable state row -> LLM response schema -> branch guard -> Telegram handoff or reservation update.
Failure checks
| Risk | What I check first |
|---|---|
| State overwrite | Same user sends two messages before Airtable update finishes; add idempotency key and last-state guard. |
| Invalid LLM shape | OpenAI response is prose or partial JSON; add schema validation and fallback branch. |
| Handoff loop | Telegram flag is set but WhatsApp continues AI replies; add explicit human_active lock. |
| Reservation ambiguity | Missing date/time/party size fields; add required-slot state and confirmation step. |
$29 deliverable
- State diagram for one failing conversation path.
- Exact node/condition likely causing the break.
- First patch slice with validation payloads.
- No API keys or customer data needed; sanitized screenshots/export are enough.