LLM Slackbot

This ia a hackathon submission for Uncharted summer hackathon 2023. The goal is to create an LLM-powered Slackbot that is able to synthesize human conversations and to provide summarization and additinal context.

Example scenarios:

• If you take a short vacation, rather than reading thorugh potentially hundreds of messages, you can just ask the bot to provide an executive summary
• If there is a conversation that is either too technical or theoretical, you can ask the bot to provide additional resources

This is a one-day hack, however due to other obligations most of the code was written within a time span of 4 hours. The code "works" but does not represent how things should be done, nor good engineering practices.

Prerequisite

• Copy env.sample.json to env.json and fill out the credentials.
• NodeJS 18

Run LLM usecases by themselves

• Copy openAI key into env.json
• Run npm install
• Run node ./llm-tests.js

Run Slackbot

• Setup and authorize a Slack App/API, this is a good guide that provides the necessary configurations for this bot.
  • https://medium.com/walmartglobaltech/create-slackbot-using-slack-bolt-api-and-node-js-a82876db012f
• Copy credentials as needed to env.json
• Invite the bot to the channel(s) on Slack
• Run node ./slackbot.js

Informal result

Using GTP-3.5 turbo model. We found that summarizations were decent and potentially useful. The "expand" command to expand current knowledge scope yielded convincing, but often incorrect results - the LLM model here is hallucinating and provided external links that goes nowhere.