Embeddings

The idea is to replace comparison by words with comparison by meaning. Substring search will never match "limescale remover" to "how to clean a kettle", because they share no words. Embeddings turn both phrases into vectors of, say, several hundred or several thousand numbers, and proximity is measured as the angle between them, the cosine similarity. The closer the vectors, the closer the meaning. The same trick applies beyond text: product listings, images and user behaviour, anything representable as a vector, can be compared by similarity.

The commercial scenarios are broader than people assume and not all of them involve chatbots. Site search that copes with synonyms and typos. Similar-product recommendations without manual category tagging. Routing support tickets by meaning rather than keywords. Deduplicating product listings when merging supplier price lists, a job that takes weeks by hand. Clustering reviews and enquiries to see what people write about most. And, of course, the retrieval layer inside a RAG system.

Three traps, all technical. First, mixing vectors from different models: they are not comparable, search starts returning noise and nothing visibly breaks. Hence the rule that changing the model means reindexing everything. Second, model quality on Russian text: many popular models are trained mostly on English and perform noticeably worse on Russian, so the choice has to be validated on your own data rather than on benchmark tables. Third, expecting vector search to replace the ordinary kind: part numbers, document IDs and exact names are still retrieved more reliably by full-text search, and the correct design is usually hybrid.

Related terms

• RAGRAG is a pattern where a language model first retrieves relevant fragments from your own knowledge base and answers from them, which delivers current data and a source citation without any fine-tuning of the model itself.
• LLMAn LLM is a large language model: a neural network trained on vast amounts of text to predict the next token, which is why it can write, summarise and answer, yet it stores no facts the way a database does and can be confidently wrong.
• AI agentAn AI agent is a program in which a language model does not merely answer but acts: it receives a goal, chooses its own tools such as search, APIs, a database or sending mail, works through steps in a loop and stops when the task is done or a limit is hit.
• Prompt engineeringPrompt engineering is the design of instructions for a language model: role, context, task, output format and constraints are written so the result is reproducible, and quality is judged against a set of test cases rather than by impression.
• Semantic coreA semantic core is an intent-grouped list of the search queries a site should be found for: collected from Wordstat, Webmaster and competitor tools, then clustered by results-page overlap so you know how many pages are needed and what each one covers.

Related services

• AI and ML developmentWe build AI that solves a defined task and pays for itself, not demos for the sake of demos. Classifiers, recommendations, text and document processing, LLM assistants, and model integration into an existing product.
• E-commerce developmentA store is not a shop window, it is an operational tool: catalogue, payments, delivery, stock and orders have to work as one. We build it so your team can run it without a developer.
• IT consulting and product auditThe most expensive mistakes in software happen before the first line of code: a misread problem, a stack chosen out of a contractor’s habit, and a specification that does not exist. Consulting exists to settle all of that before the development meter starts running: what to build, out of what, at what cost and in which order. The result is a document, not an opinion on a call.

Read more

• An AI Assistant on Your Own Knowledge Base: How RAG Works and What It CostsHow a RAG assistant differs from a button-tree bot and from an LLM with no data, where hallucinations come from and what actually reduces them, what SaaS costs against a custom build, and which month the project breaks even.
• Where AI and ML Actually Pay Off in a Business, and Where They Are Expensive HypeMachine learning pays off where there is a repeated decision, a lot of similar data and tolerance for error. Missing any one of the three means do not. Here are the use cases that work, their realistic accuracy, the running costs and how to run a four to six week pilot.