Productionizing Your Pipeline

Implementing a robust search system requires strategic productionization for scalability and performance. Understanding the critical steps and best practices makes the transition from a prototype to a production-ready solution smooth and efficient.

Choose Your Pipeline

An effective pipeline is the foundation of your search system. It’s crucial to test and evaluate it thoroughly to ensure its performance meets the users’ needs.

Have a look at this checklist and make sure you completed each of these tasks:

If you find that your pipeline still needs improvements, see Optimizing Your Pipeline.

Plan Your App

The success of your search system hinges not only on the quality of your pipeline but also on the usability of your user-facing app. When planning your app, consider the points below.

Your Data

If your pipeline operates on static data, you just need to upload your files to deepset Cloud or your VPC, and you’re set.
If your data changes frequently, set up a continuous data integration. Make sure you enable overwriting of outdated files. See Uploading Asynchronously for example scripts and the tutorial for end-to-end instructions.

Scaling

deepset Cloud takes care of the scaling for you, so you don’t need to worry about it. For deepset Cloud capabilities and benchmarks, see Scaling.

Integrate deepset Cloud with Your App

Use the Search API endpoint to add the search powered by deepset Cloud to your customer-facing app. For detailed instructions, see Use Your Pipeline in Your Target App.

Consider also setting the default pipeline to which you want to route all traffic. deepset Cloud offers a dedicated API endpoint for this. If you need to improve or change your pipeline, you can use this endpoint to seamlessly redirect traffic to another pipeline without your users noticing.

Ensure the pipeline you plan to use is set to the Production service level. This ensures proper scaling and reliability. To learn more, see Pipeline Service Levels.

Collect Feedback

Gathering user feedback helps you assess your pipeline’s performance. Ensure your users have the opportunity to let you know what they think about your app. You can use the add feedback endpoint to import all the feedback into deepset Cloud and analyze it there.

Set Expectations

Be transparent with your users about how your system works. This includes the boundaries it may have, what type of queries it can handle, where the answers come from (your documents only or the model’s general knowledge), and the like. For more guidance, see Guidelines for Onboarding Your Users.

Start Small

Consider releasing the system to a small number of users first. Their feedback will be crucial in determining if your system lives up to their expectations or if you need to improve it some more. Once the initial users are happy with the performance, you can slowly expand the system’s availability.

Iterate

Developing a search system doesn’t end when you release the app. It’s an iterative cycle of monitoring, collecting feedback, and making improvements. Embracing this iterative process will help you evolve the system in alignment with user needs and expectations.