Client project — AI-powered email support automation
Who this is for
Businesses that get repetitive customer/support emails — where the same questions get answered over and over, using information that already exists in company documents, but someone still has to read each email and type a reply by hand.
The problem
Writing accurate email replies means knowing the company's product details, policies, and past answers — and typing a fresh response every time. As the document library grows, keeping every reply consistent and up to date becomes harder to do manually.
What I built
A self-updating AI email agent built in n8n:
- Watches a Google Drive folder — any new document is automatically converted to Markdown, summarized, embedded, and added to a Pinecone knowledge base
- Watches Gmail for incoming emails
- Classifies each incoming email by type/intent
- Uses the Pinecone knowledge base (with conversation memory) so replies are grounded in the company's actual documents, not generic AI guesses
- Drafts a reply, generates a formatted HTML email template, and routes it through a review step before sending
- Automatically sends the finished reply through Gmail
Tech stack
n8n, Pinecone (vector database), OpenAI (embeddings + chat model), Google Drive API, Gmail API
The result
- The knowledge base grows on its own — adding a new document to Drive is enough to make the AI "know" it, no manual re-training
- Replies are grounded in real company documents instead of generic AI answers, reducing incorrect or inconsistent responses
- A review step keeps a human in the loop before anything is sent, balancing automation with control
