Best Practices and Tips for LLM Application Engineering - Dr Julian Seidenberg - Auckland AI Meetup

Learn from Dr Julian Seidenberg, a seasoned Artificial Intelligence Architect with two decades of experience, as he shares insights on best practices and tips for developing solutions using Large Language Model API’s.

You will learn proven tips and tricks and theory behind why these tricks work.

Many thanks to Vector Technology Solutions (VTS) for sponsoring and hosting this event!

Slides download: https://1drv.ms/b/s!AmM9o8BYCLPlg4c9j...

Resources mentioned in the closing remarks:

· More reliable structured responses from LLM's than asking for JSON: Zod Schema with LangChain https://js.langchain.com/docs/modules...

· Vector distance not getting you relevant enough documents for your RAG solution? Consider utilizing Reranker models to choose the best ones. Google search is your friend for more info on this one.

· Open source models, particularly specialized ones, can be a good alternative IMO, in niche or cost-prohibitive LLM applications. FastChat is an easy-to-use framework for getting them working on your GPU easily.

· Don't have a dataset with lots of labelled data to retest your RAG solution? Consider TruEra's generalized testing model.