An assistant that recommends the right products, and never makes them up

Context

Malvi’s Cooking offers its users a range of products and content.

With a rich offer, users do not always know what to choose. Good advice, tailored to their situation, often makes the difference.

The challenge

Malvi’s Cooking wanted to use AI to create a more personalised recommendation experience.

The risk with a standard AI assistant is well known: it answers confidently, even when it is wrong. An assistant that recommends a product that does not exist, or misdescribes an offer, loses the customer’s trust.

Our approach

Our starting rule: the assistant must only talk about what actually exists. Instead of letting the model answer from memory, we make it consult the brand’s knowledge before every answer.

What we built

The data foundation

We structured and connected the information the assistant needs, so it draws on reliable knowledge of the offer and retrieves the relevant information for each conversation.

The AI layer

We developed a conversational assistant built on a RAG (retrieval-augmented generation) architecture.

In practice, before generating a response, the system searches the knowledge base for relevant information. Every recommendation is built from the products and content that are actually available.

Results

Malvi’s Cooking has an assistant that turns its product knowledge into personalised recommendations and sales conversations.

Every recommendation stays grounded in the brand’s real offer.