An AI assistant that only says true things
A production chatbot answering customers on WhatsApp and Messenger from a verified product knowledge API.
- Client
- Bionic
- Our role
- Architecture, integration, deployment
- Stack
- OpenAI · n8n · Laravel · WhatsApp API · Messenger
The problem
A generative assistant that invents a product claim is a liability, not a feature. The brand needed automated answers that were fast and available around the clock, but could not risk the bot inventing a health claim about a product.
What we did
- 01
Built a verified knowledge API as the only source the assistant is allowed to answer from
- 02
Wired the conversation flow through automation workflows so product data, orders and escalation all stay in sync
- 03
Added an image generation pipeline for marketing content, with guards that fail loudly instead of producing empty output
- 04
Set explicit handover rules so a real person takes over the moment the conversation leaves known ground
Outcome
24/7
customer coverage
2
channels served by one knowledge base
Next projectRebuilding a storefront around what is provable