Build a Philosophy Quote Search and Generator

Philosophy with Vector Embeddings, OpenAI and Astra DB

AstraPy version

In this quickstart you will learn how to build a "philosophy quote finder & generator" using OpenAI's vector embeddings and DataStax Astra DB as the vector store for data persistence.

The basic workflow of this notebook is outlined below. You will evaluate and store the vector embeddings for a number of quotes by famous philosophers, use them to build a powerful search engine and, after that, even a generator of new quotes!

The notebook exemplifies some of the standard usage patterns of vector search -- while showing how easy is it to get started with Astra DB.

For a background on using vector search and text embeddings to build a question-answering system, please check out this excellent hands-on notebook: Question answering using embeddings.

Table of contents:

• Setup
• Create vector collection
• Connect to OpenAI
• Load quotes into the Vector Store
• Use case 1: quote search engine
• Use case 2: quote generator
• Cleanup

How it works

Indexing

Each quote is made into an embedding vector with OpenAI's Embedding. These are saved in the Vector Store for later use in searching. Some metadata, including the author's name and a few other pre-computed tags, are stored alongside, to allow for search customization.

Search

To find a quote similar to the provided search quote, the latter is made into an embedding vector on the fly, and this vector is used to query the store for similar vectors ... i.e. similar quotes that were previously indexed. The search can optionally be constrained by additional metadata ("find me quotes by Spinoza similar to this one ...").

The key point here is that "quotes similar in content" translates, in vector space, to vectors that are metrically close to each other: thus, vector similarity search effectively implements semantic similarity. This is the key reason vector embeddings are so powerful.

The sketch below tries to convey this idea. Each quote, once it's made into a vector, is a point in space. Well, in this case it's on a sphere, since OpenAI's embedding vectors, as most others, are normalized to unit length. Oh, and the sphere is actually not three-dimensional, rather 1536-dimensional!

So, in essence, a similarity search in vector space returns the vectors that are closest to the query vector:

Generation

Given a suggestion (a topic or a tentative quote), the search step is performed, and the first returned results (quotes) are fed into an LLM prompt which asks the generative model to invent a new text along the lines of the passed examples and the initial suggestion.

Setup

Install and import the necessary dependencies:

Connection parameters

Please retrieve your database credentials on your Astra dashboard (info): you will supply them momentarily.

Example values:

• API Endpoint: https://01234567-89ab-cdef-0123-456789abcdef-us-east1.apps.astra.datastax.com
• Token: AstraCS:6gBhNmsk135...

Instantiate an Astra DB client

Create vector collection

The only parameter to specify, other than the collection name, is the dimension of the vectors you'll store. Other parameters, notably the similarity metric to use for searches, are optional.

Connect to OpenAI

Set up your secret key

A test call for embeddings

Quickly check how one can get the embedding vectors for a list of input texts:

Note: the above is the syntax for OpenAI v1.0+. If using previous versions, the code to get the embeddings will look different.

Load quotes into the Vector Store

Get a dataset with the quotes. (We adapted and augmented the data from this Kaggle dataset, ready to use in this demo.)

A quick inspection:

Check the dataset size:

Write to the vector collection

You will compute the embeddings for the quotes and save them into the Vector Store, along with the text itself and the metadata you'll use later.

To optimize speed and reduce the calls, you'll perform batched calls to the embedding OpenAI service.

To store the quote objects, you will use the insert_many method of the collection (one call per batch). When preparing the documents for insertion you will choose suitable field names -- keep in mind, however, that the embedding vector must be the fixed special $vector field.

Use case 1: quote search engine

For the quote-search functionality, you need first to make the input quote into a vector, and then use it to query the store (besides handling the optional metadata into the search call, that is).

Encapsulate the search-engine functionality into a function for ease of re-use. At its core is the vector_find method of the collection:

Putting search to test

Passing just a quote:

Search restricted to an author:

Search constrained to a tag (out of those saved earlier with the quotes):

Cutting out irrelevant results

The vector similarity search generally returns the vectors that are closest to the query, even if that means results that might be somewhat irrelevant if there's nothing better.

To keep this issue under control, you can get the actual "similarity" between the query and each result, and then implement a cutoff on it, effectively discarding results that are beyond that threshold.
Tuning this threshold correctly is not an easy problem: here, we'll just show you the way.

To get a feeling on how this works, try the following query and play with the choice of quote and threshold to compare the results. Note that the similarity is returned as the special $similarity field in each result document - and it will be returned by default, unless you pass include_similarity = False to the search method.

Note (for the mathematically inclined): this value is a rescaling between zero and one of the cosine difference between the vectors, i.e. of the scalar product divided by the product of the norms of the two vectors. In other words, this is 0 for opposite-facing vectors and +1 for parallel vectors. For other measures of similarity (cosine is the default), check the metric parameter in AstraDB.create_collection and the documentation on allowed values.

Use case 2: quote generator

For this task you need another component from OpenAI, namely an LLM to generate the quote for us (based on input obtained by querying the Vector Store).

You also need a template for the prompt that will be filled for the generate-quote LLM completion task.

Like for search, this functionality is best wrapped into a handy function (which internally uses search):

Note: similar to the case of the embedding computation, the code for the Chat Completion API would be slightly different for OpenAI prior to v1.0.

Putting quote generation to test

Just passing a text (a "quote", but one can actually just suggest a topic since its vector embedding will still end up at the right place in the vector space):

Use inspiration from just a single philosopher:

Cleanup

If you want to remove all resources used for this demo, run this cell (warning: this will irreversibly delete the collection and its data!):