WhatsApp AI Knowledge Base Agent with MongoDB Vector Search
N8N
Project's Category
Intro
This one listens for WhatsApp messages, whether text, voice, image or document, and answers using GPT4o mini backed by a MongoDB knowledge base built from the team's own Google Docs.
Problems
Customers were sending questions in every format WhatsApp allows, text, voice notes, screenshots, PDFs, and the team's product documentation only lived in Google Docs nobody outside the company could search. Every question meant someone translating between the two.
Solutions
We built an n8n workflow that routes each incoming WhatsApp message by type, downloading and converting voice notes, images and documents before anything else happens. The extracted text gets embedded and matched against a MongoDB vector store holding the team's product documentation, and GPT4o mini answers using whatever it finds plus the conversation's memory. A separate manual workflow reimports and reindexes the Google Docs into MongoDB whenever the documentation changes.
Final Thoughts
Keeping indexing as its own manual workflow instead of running it on every message means updating the knowledge base costs nothing extra per conversation. Someone just reruns the import whenever the docs actually change.
WhatsApp AI Knowledge Base Agent with MongoDB Vector Search
N8N
Project's Category
Intro
This one listens for WhatsApp messages, whether text, voice, image or document, and answers using GPT4o mini backed by a MongoDB knowledge base built from the team's own Google Docs.
Problems
Customers were sending questions in every format WhatsApp allows, text, voice notes, screenshots, PDFs, and the team's product documentation only lived in Google Docs nobody outside the company could search. Every question meant someone translating between the two.
Solutions
We built an n8n workflow that routes each incoming WhatsApp message by type, downloading and converting voice notes, images and documents before anything else happens. The extracted text gets embedded and matched against a MongoDB vector store holding the team's product documentation, and GPT4o mini answers using whatever it finds plus the conversation's memory. A separate manual workflow reimports and reindexes the Google Docs into MongoDB whenever the documentation changes.
Final Thoughts
Keeping indexing as its own manual workflow instead of running it on every message means updating the knowledge base costs nothing extra per conversation. Someone just reruns the import whenever the docs actually change.
Building the automation systems that handles business processes.
Clear Communication
Reliable
Professional
Reliable Quality
Building the automation systems that handles business processes.
Clear Communication
Reliable
Professional
Reliable Quality
