Building a Neural Image Search Engine
Engineering Director, Jina AI
I am passionate about bringing machine learning live. As an Engineering Manager I like thinking about the actual user problem, diving into ML algorithms, enforcing cleaner code and setting up a good team and culture.
With the rapid growth of media and metadata in the consumer software sector, there is an evolving need for search systems to go beyond the conventional search approach and towards semantic search, or neural search. Deep learning technology provides a good base to semantically search for information. However, building a neural search system is non-trivial. In this workshop, we will show you step-by-step how you can make use of open-source tools to build a system that performs text-to-image and image-to-image search with open-source datasets.
By the end of the workshop, you will have deployed a Neural Search application for your multimodal data that serves live traffic.
The basic concepts behind Neural Search
Process your multi-/cross-modal data with DocumentArray
Train a model based on your data
Deploy a Jina Now App via CLI
Use different Indexers for different use-cases
Basic python knowledge
You will need a Jina account. Please create one upfront via
pip install jcloud (ideally in a fresh virtual environment)
During the workshop session, you will need you authentication token for the google colab notebook. You will get it via
If you want to use your own dataset during the session, please upload it upfront to your cloud provider and understand how to access it from Google Colab. You will need to share this dataset with our servers. Please don’t bring sensitive data.
We will run the entire workshop in google COLAB to avoid installation/version issues upfront.
If you want to try to run it locally, install Jina Now via pip install jina-now on the day of the workshop
We will not be able to support debugging for non-COLAB environments during the workshop.