Automate with EfeBook a Call

AI Chatbots · Jewelry

Connecting Shopify Support Questions to Answers and Human Action

I built a stateful customer-support chatbot for a Shopify jewelry retailer, connecting order context, knowledge guidance, photo evidence and structured human handoffs.

  • Shopify
  • Notion
  • Reamaze
  • Gemini
  • FastAPI

The challenge

Support questions depend on live context

Ecommerce support questions often look simple. But “Where is my order?” requires the right customer and order to be identified before current production, fulfillment and tracking information can be returned. A missing delivery, unseen design preview, incorrect product or order change each requires a different sequence.

The system needed to distinguish related intentions, remember details across messages, retrieve from the appropriate source, collect evidence when required, and recognize when the next step belonged to a support agent.

The solution

A guided path from question to action

I implemented an embeddable chat experience backed by a stateful FastAPI service. It asks for missing information before taking the next step, retrieves Shopify and retailer status context, loads Notion guidance, requests photos for relevant cases, and creates a contextual Reamaze conversation when a person must act.

AI is used selectively. Deterministic rules, validation and workflow gates handle defined business processes first. Gemini, integrated through LangChain, supports uncertain classification or confirmation and selected handoff-writing tasks without bypassing those controls.

How it works

One connected support workflow

  1. 01

    A customer starts a conversation

    The widget sends message history, a session identifier and current workflow state to the backend.

  2. 02

    The request is classified

    Deterministic intent and state rules identify the appropriate path and ask for any missing details.

  3. 03

    The system retrieves context

    Depending on the request, it reads Shopify orders and fulfillment data, retailer status records, or relevant Notion guidance.

  4. 04

    The workflow applies business rules

    It can return a status, present order choices, request the affected item or proposed change, confirm a missing delivery, or ask for photo evidence.

  5. 05

    The customer receives a next step

    Available tracking and status details are returned; defined cases continue through a guided sequence.

  6. 06

    A person takes over when needed

    The system creates a Reamaze conversation with the collected context and attachment, using stage and duplicate controls.

What I delivered

Customer experience, backend and integrations

  • A browser-based chat widget with session history, multi-turn workflow state and photo upload support.
  • A FastAPI backend with structured request and response contracts.
  • Guided workflows for order status, missing deliveries and previews, wrong or damaged products, order changes, cancellations, reorders and selected purchase enquiries.
  • Shopify order, line-item, address, fulfillment and tracking lookups.
  • Notion knowledge retrieval and Reamaze handoffs with customer context, tags and optional photo evidence.
  • Bounded Gemini and LangChain assistance for uncertain language and selected support-case text generation.
  • Vercel configuration plus targeted automated and built-in flow checks.

Reliability and oversight

Explicit controls, not unrestricted AI

Human handoff

Cases that need review or support-side action are routed to Reamaze with the information already collected.

Escalation gates

Workflow stages reduce premature or repeated cases and record when escalation has already occurred.

Validated uploads

Browser and backend checks enforce a five-megabyte limit, allowed image formats and file-signature validation.

Rate limits and authentication

IP- and session-based limits protect chat traffic, while bearer-token verification protects the direct support endpoint.

Constrained links

A guard rejects URLs that are not supported by a defined workflow or retrieved knowledge content.

Explicit failure paths

External requests use timeouts and status checks; an optional AI-summary failure does not block the underlying handoff.

These controls demonstrate implemented safeguards. They do not constitute an independent security or compliance audit.

Operational impact

A connected path from routine questions to support action

The implemented system connects Shopify, retailer order-status data, Notion and Reamaze. It supports guided order and post-purchase paths, identifies the next required detail, selects the relevant lookup, determines when photo evidence is needed, and carries conversation, order and evidence context into human review.

No client-approved performance or commercial metrics were supplied, so this case study makes no claim about time savings, ticket reduction, conversion improvement or revenue impact. These are verified implementation capabilities, not measurements of live business performance.

Discuss your support workflow

Connect your questions to real action

Build an AI chatbot around your ecommerce records, knowledge and escalation rules.

Book an AI Automation Audit