Composable-commerce storefront demo to show off the performance of Hypermode's vector search.
Example site:
Explainer Video:
When a user pauses in the search bar, Hypermode generates an embedding of the user's input and executes a vector search (HNSW or sequential) of a 10,000 item catalog in less than 200ms.
This template illustrates the power of Hypermode collections to create real-time production-grade semantic search.
This is accomplished by hosting small embedding models, and building vector indexes in memory, allowing for low latency results.
To try it out, add sample data to your collection using the upsertProducts api, which embeds & inserts texts into your index, and query the index using searchProducts.
- To deploy this code, click "use this template" in this repo.
- Go to hypermode.com/sign-up.
- Create new project
- Import this repo
Hypermode will automatically spin up the embedding model found in backend/hypermode.json file as well as generate a working GraphQL API for all functions exported from your backend/functions/index.ts file.
An example is provided, named backend/extras/hyper_toys.csv, and the corresponding python script to run it is named backend/extras/ecommerce_populate.py. Navigate to the subfolder to run them.
To install dependencies, run pip install -r requirements.txt from within that directory.
You'll need to edit the ecommerce_populate.py file with your Hypermode project URL and auth token found in your Hypermode console dashboard.
Load the data using python3 ecommerce_populate.py, and it will show you the batched inserts and time taken.
Please note: since this is inserting 10k rows sequentially, it will take ~18 minutes to embed & insert data. If you want to just try this out with a smaller dataset, feel free to shrink the csv to whatever you need.
Open up postman and create a new GraphQL request, and add the endpoint with the authorization as a bearer token. On schema introspection you will see all the functions from the template light up as APIs. Just mess around with it and see what happens!
Hypermode is a framework for building AI powered API.
In an Hypermode project, exported functions from functions/assembly/index.ts are immediately available as a scalable GraphQL API, allowing direct integration of your AI logic into your existing applications.
Each function can use AI models inferences, data connections (HTTP, GraphQL, DB, ...), collections ( a flexible vector search abstraction) and custom logic.
For more information on functions, models, connections, collections and project configuration, consult our documentation.
For writing functions in AssemblyScript, you can refer to the examples in hypermodeinc/functions-as.
To run this project locally, start by setting up your environment variables. Copy the definitions from .env.example into a new file named .env.local at the root of your project, and provide the values from your Hypermode dashboard.
Once your environment variables are configured, install the necessary dependencies with:
pnpm install
Then, start the development server using:
pnpm dev
Your app should be up and running at localhost:3000
NOTES: Make sure your environment variables are added to your Vercel project.
- HYPERMODE_API_TOKEN
- HYPERMODE_API_ENDPOINT
