This category is for people building something with a model rather than just chatting to one. It holds frameworks that connect assistants to company documents, vector databases for semantic search, platforms that track experiments and prepare training data, and an API that finds relevant web sources for an application to read.

Read more about this category

Who will maintain it?

That question narrows the list quickly. LlamaIndex links internal documents, PDFs and databases to a language model for custom chatbots, and Haystack is an open-source framework for semantic search, question answering and RAG built from pipelines. Both are libraries an engineer wires up, not products a non-developer can click through. Vector stores such as Qdrant give more control over retrieval but need hosting and tuning.

Checks that save rework

  • Can you reproduce last week's result? Weights & Biases tracks experiments and visualises metrics, which makes regressions easier to trace.
  • Where does company data go once a connector reads it, and who on the vendor side can access it?
  • If users write in Arabic, test retrieval and answers in Arabic before committing, not only in English.

Training or fine-tuning your own model is a different job, covered in Machine Learning. For storing embeddings and querying structured data, see Database.