Prompt Engineering for LLM analysis in automated reports

[john-waczak]

Within our analysis notebooks, we can easily generate strings of text with data interpolated into them. To facilitate automatic analysis of our data, we should engineer a set of prompts that can be used to supply data from a node and ask a set of questions to an LLM like ChatGPT.

ChatGPT in particular already has a set of APIs for sending prompts to their service. @davidlary do you have an API token we could use to try some prompts out?

[ashenfernando1]

We can run an LLM locally and provide it pdfs on which we can ask questions, without a need for an API token.

There also exist repos like this (which currently don't support other LLMs, thus require an OpenAI token): https://github.com/gventuri/pandas-ai, from which we can use natural language to query dataframes.

[john-waczak]

This could be useful: https://www.answer.ai/posts/2024-06-21-claudette.html

[john-waczak]

It appears Jeremy Howard has created a nice python interface to Claude from Anthropic called Claudette.