This is an example repo that explains how ReactiveSearch pipelines can be used to build an app store search backend.
Pipelines lets users define flows that are invoked based on paths. These flows can have custom stages (defined using JavaScript) or they can utilize pre built stages to override the default behaviour of a route.
In this example, we are utilizing pipelines to override the index endpoints to add an extra step that adds vector data based on a few fields and saves it to ElasticSearch.
The pipelines are automatically deployed from this repository by using the pipelines-action which uses GitHub Actions to auto deploy the pipelines and does a lot more. Read here.
As explained above, an indexing pipeline is defined that automatically adds vector data for the Name and Description fields and saves them as dense_vector in ElasticSearch. These vectors are calculated using the bert-as-service API.
Read more about this pipeline in the index directory
There are two search pipelines defined in this example.
- aNN Search Pipeline
- Text Search Pipeline
This pipeline utilizes ElasticSearch's approximate kNN (k-Nearest Neighbor) to find the search results of the query passed.
This pipeline follows the good-ol-way of doing text searches using ElasticSearchs builtin functions like match.
Read more about these pipelines in the search directory
The frontend of this app is built using ReactiveSearch components. These components are used with VueJS in this example.