No-code AI e-commerce search in under 30 minutes
Amazon have image-to-image search. Now you can too.
So, Amazon have a pretty cool feature where you can snap a picture of your favorite clothes to find similar clothes in their store. Or Tide Pods. Or teddy bears. Or Nicholas Cage pillows. You get the gist.
It’s not fair for only companies like Amazon to have that kind of tech. Image search shouldn’t just for bazillionaires with more money than God!
So, how do we build our own image-to-image search engine? And more to the point, how do we do it without pouring hours of engineering talent into it? Is there a no-code way?
That’s just what we’re going to build here, with Jina NOW. Oh, and as well as image-to-image, it’ll also do:
- Text to image
- Webcam to image
- Text to text
And you can get it all up and running in under 30 minutes, with only two CLI commands:
Step 1: Install the requirements
Only one requirement to install. In your terminal:
pip install jina-now
(of course, we recommend setting up a virtual environment first.)
Step 2: Run the program
Once you’ve done that, select some menu options, sit back, and watch your search engine spinning up! You’ll have options to:
- Download a dataset of your choice (as well as fashion there’s art, NFT’s, birds, medical imaging data, and lots more), or use your own data
- Let you select quality/speed level
- Let you deploy locally or on Google Kubernetes engine (more providers coming soon)
- Let you choose to run sandbox to do all computing on the cloud, saving you compute power. Or just run locally if that’s your thing
Based on your choices it will then:
- Encode all the data using CLIP
- Fine-tune the CLIP model with Finetuner for your specific dataset
- Deploy the Jina Flow to your machine or Kubernetes cluster with RESTful endpoints
- Provide a browser front-end for you to search
And as for the “under 30 minutes” part? I timed it using
time jina-now start :
jina-now start 606.35s user 17.89s system 37% cpu 27:39.99 total
See? Time to spare!
This is great as a demo of the power of neural search. But if you really want to get your hands dirty, check out a dedicated fashion search demo, which also offers:
- Easier self-hosting via Docker Compose.
- Faceted search capabilities via tag filtering, price ranges, etc.
The repo is totally open for you to clone, fork, and modify as you wish.
GitHub - jina-ai/example-multimodal-fashion-search: Input text or image, get back matching image…
Multimodal search lets you use one type of data (in this case, text) to search another type of data (in this case…
Learn more about Jina NOW
You can also read more in the release blog post:
Jina NOW → The one-liner Neural Search
Jina NOW is the first-ever nocode solution that lets you create and deploy multimodal neural search in a matter of…
- Jina NOW — no-code image search
- Jina Learning Bootcamp — learn the Jina ecosystem
- More complex fashion search example with faceted search