📦 Semantic Search with Typesense + Instantsearch.js

This is a demo that shows how you can use Typesense's vector search feature, to build a semantic search experience.

NOTE: ⚠️ This demo uses Typesense version 0.24.0 which did not have built-in embedding generation. But in v0.25.0, we've added built-in emebedding generation, which makes it even more easy to build semantic search out of the box, without the need for an external embedding service. Here's a newer demo that takes advantage of this feature: https://github.com/typesense/showcase-hn-comments-semantic-search.

Learn more about Vector Search here: https://typesense.org/docs/0.25.0/api/vector-search.html.

Tech Stack

The app was built using the Typesense Adapter for InstantSearch.js.

Repo structure

• src/ and index.html - contain the frontend UI components.
• scripts/indexer - contains the script to index the book data into Typesense.
• scripts/data - contains a small sample subset of products.
• server - contains the API server that generates embeddings given a query.

How to

Index sample dataset:

npm install
npm run typesenseServer

ln -s .env.development .env

npm run indexer

Start Embedding API server:

cd server

pip install -r requirements.txt

uvicorn main:app --reload

Start FE app:

npm start

Open http://localhost:3001 to see the app.

Credits

The dataset used in this showcase is from Algolia's public set of datasets listed here: https://github.com/algolia/datasets