Skip to content

Latest commit

 

History

History
113 lines (73 loc) · 2.6 KB

readme.md

File metadata and controls

113 lines (73 loc) · 2.6 KB
Raw

Monster menu - Demo AI Heroes 2023

Source of the demo of the talk Integrate LLM in your applications 101 of AI Heroes 2023 .

Slides https://www.slideshare.net/marcodenittis/integrate-llm-in-your-applications-101

The demo is a simple bash application integrated with OpenAI in order to build a chief assistant bot for a fictional restaurant.

This demo is tested on MacOS and should work on any posix system.

Requirements

  • A valid OpenAI API key in the environment variable OPENAI_API_KEY
  • jq installed
  • curl installed
  • docker installed to run redis
  • gcc installed in order to build the tool to convert the embeddings in binary/utf8 format

Structure

Project folders

  • assets: intermediate files used by the demo
  • data: demo data as recipes
  • scripts: bash scripts
  • templates: templates from the prompts

Runtime folders

  • .cache: cache of embeddings
  • .data: docker volume for redis and other runtime data
  • sessions: markdown logs of the operations

Scripts:

session: manages the current log session (in the ./session folder)

# show the current session id
session

# start a new session
session new

ask: main script to interact with the openai api

Samples:

# simple prompt with no template
ask what could I offer to a vampire guest

# prompt using the template 'templates/prompts-recipe.txt'
ask -p recipes what could I offer to a vampire guest

embeddings: (alias emb, embed) generates an embedding for a given text

# simple embedding with full vector output
emb what could I offer to a vampire guest

# token count
emb -t what could I offer to a vampire guest

# embedding summary (token + first 10 values of vector)
emb -s what could I offer to a vampire guest

# embedding json response 
emb -j what could I offer to a vampire guest

rd: redis interface

# search via semantic search
rd search vampire

# add vector record to redis
rd embed what could I offer to a vampire guest

# scan a folder and add all the files contents with vector embeddings to redis 
rd embed-folder ./data/recipes

# create a vector search index
rd create-index 

# empty the redis database
rd clear-all

Usage

Before start you should run the setup command in order to initialize all the tools and set the alias.

. ./setup.sh
  • startups the redis container
  • builds the tool to convert the embeddings in binary/utf8 format (float-to-blob)
  • checks the OpenAI API key