LLM under the hood: choosing effective technical solutions for AI assistants

This course is dedicated to how to rationally select technical solutions for LLM-based product tasks in various application areas.

I will present you with a set of tools, practical cases and patterns, using which you will be able to:

• select optimal solutions for various LLM-related tasks
• modify these solutions for the specifics of your area.

We will study how to develop solutions for LLM-based products, saving time, resources and effort.

The training methodology is built on my consulting approach and successful examples of artificial intelligence implementation.

Who is it for?

This course is intended for professionals creating products involving LLM both individually and as part of a team (for example, in large companies). It will be useful for:

• Engineers developing LLM-based solutions in various domains.
• Tech leads and CTOs who need to quickly find optimal solutions using LLM for various tasks.
• Product managers leading the implementation of LLM-based solutions.
• Founders of companies developing products for the market based on LLM or adapting LLM for business processes.

The course does not teach how to work with frameworks, connect to LLM, or index documents.

Course structure

Module 1: Basics

In this module, we will study standard approaches to implementation LLM and see their limitations by looking at a common example problem. We will get acquainted with mental models and heuristics for identifying the causes of limitations and eliminating them. This knowledge could have saved me 2-3 months of work in the past.

Some of the material overlaps with our previously conducted webinars, but the course presents more information and in more detail.

If you were unable to attend the webinars, now is a great opportunity to listen to them - and much more.

Module 2: Cases and Patterns

We will study recurring architectural patterns using the example of successful AI projects. The pattern library includes: Query Expansion, Dedicated Agent, Router, Learn from Feedback, Knowledge Base and other patterns that can be adapted to the tasks at hand. Also included is Checklist + Custom Chain of Thought.

For all patterns, we will analyze the features of their application and examples of successful projects. I will demonstrate how to classify and generalize tasks for more effective solution. This will save time in future projects, consciously applying successful experience, rather than creating from scratch.