Table of Contents
Before we dive into all of the different pieces of building an LLM application, we recommend that you read through our guides on Large Language Models to get a solid foundational understanding of the core technologies we are working with, and then continue to the rest of this guide!
What is an LLM? (Large Language Model)
Architecting LLM Systems for Chatbot Services & Intelligent Company Databases
We recommend reading the development overview for a high-level overview of each stage of development of LLM projects:
It’s important to get a lay of the land in order to inform what approach you’re going to take. This is similar to the data exploration and analysis stage in any data science or AI initiative!
LLM Data Exploration Jumpstart
Here, we go over some of the core principles to follow when actually prompting the LLM.
In order to provide the LLM with the most relevant and up to date information for your use case, you’ll have to conduct a retrieval step to serve it that information. Although a lot of tools offer decent performance out of the box, it can get pretty complex going from proof of concept to production in this area.
Retrieval Augmented Generation - Best Practices
Document Chunking - Choosing Settings
It’s critical to establish a standardized and principled method of evaluating your application’s performance. This is even more true in the context of LLMs, which have characteristically nondeterministic outputs. Make sure to evaluate your performance not just at the end, but throughout the whole timeline of your project!
LLM Test and Evaluation Best Practices
Once the system has been built, tested, and fully set up for a production use case, it’s time to deploy! At PressW, we recommend using tools that allow the system to be deployed to any cloud (e.g. containerization) or even self-hosting if that’s your thing. When deploying to a cloud, we suggest using Google Cloud for ease of development. If you need higher levels of security or to use the enterprise versions of OpenAI’s GPT, we recommend setting up your system in Microsoft Azure.
For an overview on the Google Cloud services we primarily make use of, check out: Google Cloud Guides
Detailed Guides
Creating a Google Cloud Project
More details on other platforms soon!
<aside> Here’s a list of fresh new guides that our team is working on right now. As soon as they’re done they’ll be moved into their respective categories!
</aside>