An AI support bot that answers from your own content

The challenge

Online stores field the same handful of questions all day — where's my order, what's your return policy, how do I pick a size. It burns support hours, slows replies, and gets worse outside business hours. Off-the-shelf chatbots tend to either deflect nothing useful or confidently make things up.

What I built

An AI support assistant grounded only in the store's own content. It answers instantly in the brand's voice, as a website chat widget or on WhatsApp / Messenger — and it's built to be safe to put in front of real customers.

• Grounded answers: retrieves from the store's docs/FAQs (RAG) — it can't invent a policy that isn't there.
• Verifiable sources: every answer links to the exact source it used — one click to check.
• Knows its limits: when it isn't confident, it hands off to a human instead of guessing.

How it works

• Ingests the store's help docs, FAQs, and product info into a retrieval layer.
• On each question, it pulls the most relevant content and answers from that context only.
• Cites the source documents alongside the answer.
• Routes low-confidence or out-of-scope questions to the human team — no hallucinated promises.

The impact

• Deflects roughly 40% of routine tickets, around the clock.
• Faster responses; the support team focuses on the cases that actually need a person.
• Safe by design — grounded, sourced, and honest about what it doesn't know.
• Live on a real store in about two weeks, trained on your own content.

Tech

Shown here as a working demo on a sample DTC brand (“Northwind Threads”). The same system drops onto your real content and channels.