Arize AI debuted capabilities for troubleshooting large language models (LLMs).

Arize's new prompt engineering workflows, including a new prompt playground, enables teams to find prompt templates that need to be improved, iterate on them in real time, and verify improved LLM outputs.

Prompt analysis is an important component in troubleshooting an LLM's performance. Often, LLM performance can be improved simply by testing different prompt templates, or iterating on one to achieve better responses.

With these new workflows, teams can:

- Uncover responses with poor user feedback or evaluation scores

- Identify the template associated with poor responses

- Iterate on the existing prompt template

- Compare responses across prompt templates in a prompt playground

Arize is also launching additional search and retrieval workflows to help teams using retrieval augmented generation (RAG) troubleshoot where and how the retrieval needs to be improved. These new workflows will help teams identify where they may need to add additional context into their knowledge base (or vector database), when the retrieval didn't retrieve the most relevant information, and ultimately understand why their LLM may have hallucinated or generated sub-optimal responses.

"Building LLM-powered systems that responsibly work in the real-world is still too difficult today," said Aparna Dhinakaran, Co-Founder and Chief Product Officer of Arize. "These industry-first prompt engineering and RAG workflows will help teams get to value and resolve issues faster, ultimately improving outcomes and proving the value of generative AI and foundation models across industries."