AI Support Assistant — Northwind Outfitters
A demo store I built to show exactly what yours would do: a storefront assistant that answers customers from the store's own policies, shows its sources, and hands off to a human when it doesn't know.
Store owners drown in the same support questions — shipping times, return rules, order status — while the answers already sit on their policy pages. Generic chatbots make it worse: they answer confidently even when they're wrong.
The question this demo answers: can an assistant be trusted to speak for a store — accurately, with receipts, and with the humility to hand off when it doesn't know?
A complete demo storefront — Northwind Outfitters, a fictional outdoor-gear store on sample data — with an embedded chat assistant trained on its shipping, returns, warranty, and product-care policies.
The assistant answers in plain language, cites the exact policy passage under every answer, and escalates to a human when confidence is low.
Store policies are chunked and embedded into ChromaDB. On each question, LangChain retrieves the most relevant passages and the model — Llama 3.3 on Groq — answers only from that retrieved context, which is what keeps it from making things up.
A FastAPI backend serves the whole flow, and the widget is a single embeddable script tag that drops into any existing storefront.
A working, honest demo of grounded retrieval: answers with sources, graceful escalation, and zero invented facts. It's the proof behind the offer — see the AI Support Assistant page for how it becomes yours.
Live order-status lookups against real store APIs, proactive shipping notifications, and return initiation — the Connected and Automated tiers of the offer.