Viska

Viska is a personal project about conversational commerce: using intelligent chat to make product discovery and the customer experience in online stores better.

The idea is that a visitor should be able to ask a question in their own words — what would suit a beginner, will this fit my setup, what’s the difference between these two? — and get an answer that is specific, accurate and useful for that particular store.

Online stores are good at showing products and less good at answering questions about them. Filters and menus assume the customer already knows which attributes matter. Search assumes they know the right words.

The answers usually exist somewhere — in product data, specifications, buying guides, policies and support articles — but they are scattered, and finding and combining them is left to the customer.

Viska has been an area of practical experimentation with language models, retrieval-augmented generation, product knowledge and conversational interfaces. The general approach:

1. Gather the store’s own knowledge — product data, descriptions and the guides and policies around them — and prepare it so that relevant pieces can be found quickly.
2. Retrieve before answering. For each question, find the products and passages that actually bear on it.
3. Let the model compose, not invent. The language model writes the answer from what was retrieved, so it can point to real products and sources rather than relying on general knowledge.
4. Know the limits. When the store’s knowledge doesn’t cover a question, the honest answer is to say so.

The ordering matters for trust. An assistant that can show where an answer came from is easier to check, easier to correct, and easier for a business to stand behind.

The interesting questions are less about whether a model can talk about products — it can — and more about whether it can be right about these products:

• How should product knowledge be represented so the right pieces can be retrieved?
• When is a conversation better than search, and when is it just slower?
• How does an assistant stay grounded in the catalogue instead of sounding plausible?
• What does a helpful answer look like for someone who doesn’t yet know what they need?

The same principle — answer from approved sources, show them, admit gaps — is what this site’s Ask this site feature is built on.

Viska is a concept and personal project: an ongoing area of experimentation rather than a launched product. Details are kept general here on purpose; nothing on this page describes a customer, a deployment or commercial results.