An AI shopping assistant you can actually try
Most AI customer support is bad in a predictable way. It invents policies that do not exist. It promises tracking it cannot deliver. It bends under polite pressure. It breaks character the moment someone asks it to. We built a customer-facing AI for a fictional fashion brand called Veloura because we wanted to show what the opposite looks like, in code, in production, with the URL right here. The point of this post is not to talk about it. It is to get you to open it and poke at it.
By Ivaylo Tsvetkov, Co-Founder
Here is one you can try right now
The live demo is at https://forgingapps.com/en/demo/veloura-shop. Open it on a desktop or a phone, click the chat bubble in the lower-right, and start asking questions. The suggested prompts will get you moving -- "what hoodies do you have", "show me items on sale", "what sizes does the Heavyweight Hoodie come in" -- but the more interesting test is to ask it something off-script. Ask whether you can return a sweater you wore once. Ask whether it ships to Antarctica. Ask whether there is a VIP discount for loyal customers. You can also try to make it misbehave: ask it to write a poem about a hoodie, or to break character and recommend a competing brand. The replies tell you more than any marketing page will.
What it knows
The assistant is built on top of a real product catalogue and a written policy document. It can recommend across seventeen items in five categories -- hoodies and sweatshirts, t-shirts and tops, pants and joggers, jackets and outerwear, accessories. It knows the shipping policy: standard EU delivery in three to five business days, express in one to two, free standard shipping on orders over €60. It knows the return policy: thirty days, unworn, unwashed, with original tags. It knows that exchanges are limited to one size swap per order if the stock is there. It can guide on sizing across fits. None of this is generic AI knowledge. It is the brand's own content, given to the assistant as a closed source of truth so it can answer customers without making things up.
What it deliberately will not do
Just as important as the in-scope list is the out-of-scope one. The assistant will not track an order, process a refund, change an account, or run a payment. Those flows stay on the rails the team controls. This is not a limitation we hide; it is the design. A real production assistant for a small business should have boundaries -- boundaries that are easier to defend than to apologise for later. If a customer asks for one of these flows, the assistant says so and offers an escalation path with the full conversation context preserved. No dead ends. No lost threads. The customer is not left talking to a wall.
What an honest exchange looks like
Try this in the demo. Ask: "Can I return a jacket I bought two months ago? It still has tags." The standard return window is thirty days. A naive assistant either agrees politely to keep the customer happy, or refuses bluntly and ends the conversation. The Veloura assistant does neither. It answers the question against the policy -- two months is outside the window -- and then offers to escalate to the support team for a case-by-case review. The customer gets a clear answer, the brand keeps its policy intact, and the human team sees only the cases that actually need their attention. Now try the other end. Ask: "What is the cheapest hoodie you have right now?" The assistant returns the right product -- the Heavyweight Hoodie on sale at €48 -- and offers to walk through colours and sizes. No invented discount, no over-promised stock. Just an answer.
How it stays in character
The architecture under the hood is straightforward by design. A system prompt sets the persona and the policy. A retrieval layer connects the assistant to the product catalogue. A thin guard sits in front of the model and handles the predictable routes -- show me hoodies, browse the catalogue -- without burning a model call. Adversarial inputs go to the model where the policy refuses them. The result is a conversation that feels grounded because it is: the assistant is not improvising from training data; it is reading from the same content the brand wrote down. When a customer tries to push it off-character, the same canonical refusal comes back, every time. It is not clever. It is consistent, which is more useful.
Why this matters for a small business
The expensive mistake we see most often is starting an AI project with the most impressive demo in the room. A free-roaming agent. A general-purpose copilot. A model wired directly to inventory and payment. That is the path with the most variance -- high ceiling, very high floor of risk -- and it is the wrong first move for most businesses. The better first move is what the Veloura demo shows: a bounded assistant, connected to approved content, with clear escalation. It deflects the repetitive questions that eat your support time. It stays consistent across hundreds of conversations. It hands off the ones that need a human. And when something changes -- a new return rule, a new sale, a discontinued product -- you update the source content once, not the assistant. That is the kind of AI project that pays back in the first quarter, not in year three. The Veloura demo is the same architecture we ship to clients. The AI Chat Assistant offering starts at €1,500 across three tiers, with the first four months of managed Care included. We test every build against real edge cases and adversarial queries before anything goes live -- the same testing methodology we use on our own demos.
Try it, then tell us
The demo is at https://forgingapps.com/en/demo/veloura-shop. Open the chat, ask anything, and see how it answers. If you have ever wished your shop or service had something like this -- answering the same fifteen questions, twenty-four hours a day, without losing your tone -- we would like to hear what you would build. The contact link is on the demo page, and so are we. The hard part is usually the first foot in the door. This is meant to make that foot easier to put down.
Want to discuss how this applies to your business? Book a free call.
Ready to add AI to your business?
We help businesses identify, design, and deploy AI systems that actually work. Book a free discovery call and see what's possible.
Book a free call →Related Posts
View all posts →AI Chatbots for E-commerce Support: What to Automate First
A practical framework for e-commerce teams deciding what an AI chatbot should handle first: repetitive questions, product discovery, policies, and human handoff.
AI Security Is Now a Buying Issue
Anthropic's Mythos preview and an OpenAI tooling incident say the same thing: buying AI means vetting permissions, connectors, and blast radius, not just the model.
Why a Local Dev Studio Beats a Freelancer for Custom Software
Freelancers are cheap. Agencies are expensive. Boutique dev studios offer something better: senior expertise, direct accountability, and no outsourcing.