Are there any guides on customising the UI to use a custom elastic search document store?

[danielduckworth]

Hi, I've been working my way through the tutorials and have set up an elastic search docker container instance (with a persistent volume) and a document store using some FAQ data (Tutorial 4).

Are there any guides on adapting the streamlit UI app to make use of this instance instead of the default one? It would be useful to demonstrate some use-cases to my colleagues using our data.

[brandenchan]

Hi @danielduckworth, a Streamlit app for FAQ would be really cool! As I see it, this should be possible. I just want to set the scene here a bit first. So a Haystack pipeline can be deployed by wrapping it in a REST API application. The Streamlit UI communicates with the REST API in order to fire requests at the Haystack pipeline. By calling docker-compose up you can start up an ES instance, the Haystack REST API app and also the UI. I would highly recommend looking through our REST API documentation which gives a detailed guide on how to interact with these components. But I can give a high level overview of the steps that I think we need to take:

1. Ensure that your ES instance connects to Haystack
2. Ensure that your FAQ pipeline is being used by the Haystack REST API
3. Adjust the Streamlit UI so that it will show the right FAQ data

For 1) if you already have a running ES instance, it would be good just to check that your instance doesn't get clash with or shutdown due to the ES instance that is created by docker-compose up.

For 2) you will need a yaml version of your FAQ pipeline. For this, you will need to use the Pipeline.save_to_yaml() function. This yaml file will then need to be moved to haystack/rest_api/pipeline/pipelines.haystack-pipeline.yml. You will also need to comment / uncomment the lines in docker-compose.yml (or docker-compose-gpu.yml) so that you are no longer using the default haystack-api image but rather your own custom pipeline.

For 3) I am not so familiar with how to make changes to Streamlit. But let us know when you get to this point. I believe that @TuanaCelik should be able to help you with this step.

I hope this makes some sense! Please let us know if you need any help at any point.

[danielduckworth]

Excellent! Thanks for the tips. I'll have a crack over the next week or so.