🔎Retrieval & Search
RAG, embeddings and AI search · 37 terms
Embeddings
8How text becomes a list of numbers, and how to tell which lists are close.
A list of numbers that stands for what a piece of text means.
A model that reads text and gives back a list of numbers.
How many numbers each embedding holds. More can mean finer meaning, but more cost.
A list of numbers with a value in each slot. It holds meaning, not exact words.
A long list of numbers, mostly zeros. Each slot stands for one word.
A score that says how alike two vectors are, by their angle.
Finding the stored vectors closest to the one you ask with.
A fast search for close vectors that may miss a few of the very best.
Vector Databases
10Where the number lists are kept, and how they are found again fast.
A place to keep embeddings and find the closest ones fast.
A vector database you rent online. There is nothing to set up or run.
A free add-on for Postgres. It lets the database store and search vectors.
A free vector database that anyone can use. You can start in minutes.
A free vector database. It is built for fast search with filters.
A free vector database. It mixes search by meaning and by words.
A map that finds close vectors fast, so you do not look at all of them.
A fast vector index made of layers of linked points.
A free code tool for fast vector search. It goes inside your own program.
A search that only looks at rows with the right tags, like a date.
AI Search
8The ways to find the right text for a question, by its words or by its meaning.
Search that finds text by what it means, not by its exact words.
Search that finds pages with the exact words you typed.
The standard way to rank pages in keyword search.
A list of each word and the pages that hold it. It makes word search fast.
A word search and a meaning search run at once, with the two lists joined.
A way to join ranked lists. Each place earns points.
A second, slow look at the top hits. It puts them in a better order.
Facts stored as things and the links between them, like a map you can follow.
RAG Systems
11How a model is handed the right text before it answers, so it can stick to the facts.
First find the right pages, then let the model answer from them.
Tying a model's answer to trusted sources, so each claim can be checked.
Retrieval Pipeline
6The steps from a pile of files to the few chunks a model reads: parse, chunk, search and pick.
The set of files an AI looks through. Then it gives its answer.
Turning PDFs, slides and pages into clean text before they are chunked.
Cutting a long file into small bits. Then search can find the right bit.
The part of a RAG system that finds the best chunks for a question.
A vague question is made clear first. Then the search runs.
Returning only the k best matches from a search. You choose the k.
Retrieval Patterns
3Ways to make retrieval smarter than one plain search, from notes on each chunk to graphs and agents.
Each chunk gets a short note that says where it came from.
RAG where the model does its own search, a few times, then answers.
RAG that makes a map of facts from your files. Then it searches that map too.