[INTELLIGENCE LAYER] Shopify RAG Chatbot

AI Layer

Embedded where it saves hours, not where it makes screenshots.

The AI layer is the operating intelligence inside the system — the part that reads the catalogue, the spec sheets, the policies, the past orders, and surfaces them at the right desk at the right time. It is not a chatbot in the corner of the storefront. It is the tooling sales, ops, and CS reach for when a question needs an answer grounded in this brand's actual data, not a model's recollection.

LAYER03/06of the system we build
SURFACES04capabilities in this layer
STACK05technologies on the bench
SOLD ASSYSTEMnever as a single service
N°01The layer

What ships in this layer.

4 surfaces, all built as part of the same system. None of them ship alone — they are the parts that make this layer work inside the whole.

  1. 01SPEC-BOT (RAG)

    Grounded answers over spec sheets and PDPs

    Embeddings over the brand's actual specifications, with citations to source. The question "what is the slip rating on this porcelain at wet install?" gets answered from the spec, not from a guess.

  2. 02ADMIN ASSISTANTS

    Quote summaries · customer-history pulls

    Inside the Shopify admin: one-click summaries of a draft quote, last-order context on the customer card, a digest of the open trade pipeline. Sales sees the brief before the call, not after.

  3. 03POLICY-GROUNDED CS

    Shipping · returns · trade answers, cited

    The customer-service team's first draft, anchored to the policy document. Every answer carries the clause it was drawn from; the rep stays in the loop, the cost of the reply goes down.

  4. 04INTERNAL TOOLS

    Extraction · classification · drafting

    Pull line items from a contractor's PDF, classify an inbound trade application, draft a fabrication brief from the order — the tasks that used to live in a junior's inbox.

Why this layer is part of the build.

aterial brands accumulate decades of knowledge — slab dimensions, finish behaviour, freight handling, install instructions, return policies — and most of it sits in PDFs, in folders, in someone's head. Without an intelligence layer, every quote, every CS reply, every admin task pays the cost of finding that knowledge again. We build the AI layer so the system remembers — and so the people inside it stop being human search engines.

WHAT WE ARE NOT

We do not take this layer as a retrofit on a storefront we did not build. If that is the work, the right call is an agency. We build the system the layer lives inside.

N°03Technical bench

How this layer is actually built.

The stack is on the bench, the wiring is documented, and the instance shipped with your system is yours to extend after handover.

ON THE BENCH
  • 01Anthropic
  • 02Embeddings
  • 03RAG
  • 04Admin tools
  • 05Policy index
5 layers · all yours after handover

Anthropic Claude as the reasoning engine, with embeddings via Voyage or OpenAI for retrieval. RAG indexes built from the brand's spec sheets, policies, and historical order data, refreshed on a cadence tied to source updates. Admin assistants delivered as Shopify embedded app surfaces. Every answer is grounded to a cited source — hallucinations are a shipping defect, not an acceptable trade-off.

spec rows, indexed
12 k
every answer cited
Grounded
hallucinations YTD
0
N°04Where it lives in the system

Inside the system you would ship.

The layer does not stand alone. These are the standing systems most often built on top of it and the adjacent layers it interlocks with.

SYSTEMS BUILT ON THIS LAYER

N°09

Quote-to-Cart Flow

$15K+avg quote transaction

Quote request → sales builds draft cart with custom freight → single-link checkout. Replaces email-PayPal ping-pong.

N°07

Trade Portal

20–40%of revenue, trade buyers

Gated B2B experience — verified accounts, net pricing, memo samples, branded tear-sheet builder.

N°08

Production Tracker

60–70%fewer CS inquiries

Domino's-style status tracker for custom 8–14 week production. Stage emails + portal + workshop photos.

ADJACENT LAYERS · INTERLOCK

N°02Operations AutomationWhere the work actually breaks — between systems.
N°04MarketingDiscovery, demand, and lifecycle — engineered into the system.

[N°05] Common questions

The questions this layer answers.

How the ai layer layer behaves in production — the trade-offs we have already taken and the ones we re-open per build.

  1. Is the AI layer just a chatbot?

    No. The AI layer is a set of operational surfaces — retrieval, product description generation, slab tagging, internal Q&A — that run against your catalogue data. A chatbot is one optional surface, not the layer.

  2. Which models do you build on?

    Claude and OpenAI primary, open-weight models where local inference and cost shape demand it. Choice of model is downstream of the surface, not the other way around.

  3. Does the AI write the entire product description?

    It writes the long-tail catalogue body against your brand voice. The hero copy on top-tier products is still written by the brand.

  4. Where does the retrieval data live?

    In a vector store we run for you, sourced from your catalogue, content, and where useful, trade documentation. We do not pipe customer data into third-party training.

N°06Next step

You want this layer? You want the whole system.

We do not take this layer as a one-off. We build from zero and grow with the brand. Every engagement begins with a paid discovery — fit before contract.

N°AOPEN A BRIEF

Build the whole system

Paid discovery sets the shape. We map your case across all six layers and ship the system the intelligence layer lives inside.

N°BREAD THE PRACTICE

How the work runs

Six chapters on how the system gets built — origin, journey, discipline, outcomes, handover. The story behind the layers.