Introduction

In this starter you will find the basic elements to build a genai application.

Langchain is the most used framework to build GenAI application Unstructured is a lib to get the data from various sources

An embedding is a way to represent a text in a vector space. It is a way to represent the meaning of a text. A similarity search is a way to find the most similar text to a given text. (Fruit is similar to apple, banana, etc.)

Installation

The project use poetry and pyenv to manage the dependencies and the python version.

Install with poetry install and activate the virtual environment with poetry shell

Export the openai api key with export OPENAI_API_KEY=your_api_key

Test it with : python src/1_call_model.py

Launch service

docker compose -f docker-compose-pgvector.yml up Beware if you already have postgres running you might have to change the port in docker-compose-pgvector.yml

5432:5432 -> 5433:5432

2_create_vector_store.py : set port to 5433

3_search.py : set port to 5433

Semantic search - RAG

Read data from files, create embeddings, Store them in postgresql

Test it with : python src/2_create_vector_store.py

Make a similarity search with them

Test it with : python src/3_search.py

GENAI TOOL - Function calling

Create a tool - define a function

See src/tools.py

Use a tool with OpenAI - Send your function definition to openai - Call the function to get your result

Test it with : python src/4_function_calling.py