Reka describes making chart images searchable by first translating their contents into text. Its September 2024 MongoDB Atlas walkthrough sends charts to a model that can process images, stores the resulting Markdown in the database, and indexes it for keyword search. An alternative generates descriptive text about each chart and searches those descriptions. The model performs the visual interpretation; the database searches the words produced from it. This gives a text-based retrieval system a way to find relevant charts.
Terms in this piece · Glossary
RAG — Retrieval-augmented generation — fetching relevant documents first and pasting them into the model's context so it answers from your data instead of memory.
embedding — A list of numbers representing a piece of text's meaning, so that similar meanings end up numerically close and can be searched.
