Retrieval-augmented generation (RAG) is a pattern that hands a language model relevant passages from your own documents at question time. Its answers then rest on that material, not only on what it learnt in training.
Semantic search matches results by meaning rather than exact keywords, usually by comparing vector embeddings of the query with embeddings of the content. A support portal can return a refund policy article for the query "get my money back" even when those words never appear in it.
A vector database stores embeddings and quickly finds the ones most similar to a query vector, usually alongside metadata filters. It is the common retrieval layer behind semantic search, recommendations and RAG.
An AI engineer builds software products and features on top of AI models, especially large language models, using prompting, retrieval, tool use and evaluation. Companies hire one to turn a promising AI demo into a dependable product.
A context window is the maximum amount of text, counted in tokens, that a language model can take into account in one request. It covers the instructions, the conversation, any retrieved documents and the reply.
Data poisoning is an attack that corrupts the data used to train, fine-tune or ground an AI model. The model then learns hidden flaws, biases or backdoors chosen by the attacker.
Embeddings are lists of numbers that represent the meaning of text, images or other data, arranged so that similar items sit close together. They let software search, group and compare by meaning instead of exact words.
A clear proven process, that delivers
You've seen our ideas - now go deeper.
Discover more insights, guides, and engineering practice.