[AI] LangChain Explained - Providing context to LLMs

As commented earlier this year, I was inspired by Alejandro AO and its Python AI projects.

I commented the PDF one here →

But today, it is the time for CSV’s and with Databases!

You guessed it.

We will be using LangChain as RAG framework to provide the CSV context to the LLM.

LangChain is an alternative to PandasAI.

Chat with CSV with LangChain

You will need to have:

• Python installed and a virtual environment ready.
• The OpenAI API keys - https://platform.openai.com/api-keys

In a nutshell, these are the python libraries you will need:

#pip install langchain python-dotenv streamlit openai 
#but lets install them relialy with the requirements.txt file

Thats why the original Alejandro AO Project has the requirements.txt file specifying the versions.

Its a great practice to install python dependencies with:

pip install -r requirements.txt

#which it is equivalent to this way below
#pip install langchain==0.0.208 python-dotenv==1.0.0 streamlit==1.25.0 openai==0.27.0 transformers==4.28.1 sentence_transformers==2.2.2 torch==2.0.1 tensorflow==2.12.0

I just removed most of them as the original file seemed to have many more than required.

Remember to place your OpenAI API key in the .env file.

And run the Streamlit App:

streamlit run main.py

A browser will open with: localhost:8501

Understanding CSV with LangChain App

This is the general workflow of the Python app:

graph TD
    A[Load environment variables with dotenv] --> B{Check if OPENAI_API_KEY is set?}
    B -- No --> C[Exit program]
    B -- Yes --> D[Set Streamlit page config and header]
    D --> E[User uploads CSV file]
    E --> F{CSV file uploaded?}
    F -- No --> G[Wait for CSV upload]
    F -- Yes --> H[Create CSV agent using Langchain and OpenAI]
    H --> I[User inputs question]
    I --> J{Is a question provided?}
    J -- No --> K[Wait for question]
    J -- Yes --> L[Process the question with agent]
    L --> M[Display result using Streamlit]

Chat with a DB with LangChain

It’s not that SQL is impossible to learn.

But why not using AI as leverage to help us with D&A tasks?

And it is possible thanks to LangChain with Python and the great work of AlejandroAO

Database Setup

sudo apt update
sudo apt install mysql-server -y
sudo systemctl start mysql
sudo systemctl status mysql
sudo systemctl enable mysql #start at boot

sudo mysql -u root -p

#mysql --version #mysql  Ver 8.0.40-0ubuntu0.24.04.1 for Linux on x86_64 ((Ubuntu))

Now that MYSQL is installed, we can load the sample DB.

CREATE DATABASE chinook;
USE chinook;
SOURCE chinook.sql; --or the name of your SQL file to load the database
SOURCE Chinook_MySql.sql; --or the name of your SQL file to load the database

SHOW TABLES;
SELECT * FROM Album LIMIT 10;
DESCRIBE Artist;
SHOW FULL TABLES;

You can see that the DB is loaded and we can see its Schema.

SELECT 
    TABLE_NAME AS `Table`, 
    COLUMN_NAME AS `Column`, 
    DATA_TYPE AS `Data Type`, 
    CHARACTER_MAXIMUM_LENGTH AS `Max Length`, 
    IS_NULLABLE AS `Nullable`, 
    COLUMN_KEY AS `Key`, 
    COLUMN_DEFAULT AS `Default Value`
FROM 
    INFORMATION_SCHEMA.COLUMNS
WHERE 
    TABLE_SCHEMA = 'Chinook'
ORDER BY 
    TABLE_NAME, ORDINAL_POSITION;

Actually, we can make already a MermaidJS ER diagram to represent it:

erDiagram
    Album {
        int AlbumId PK
        varchar Title
        int ArtistId FK
    }
    Artist {
        int ArtistId PK
        varchar Name
    }
    Customer {
        int CustomerId PK
        varchar FirstName
        varchar LastName
        varchar Company
        varchar Address
        varchar City
        varchar State
        varchar Country
        varchar PostalCode
        varchar Phone
        varchar Fax
        varchar Email
        int SupportRepId FK
    }
    Employee {
        int EmployeeId PK
        varchar LastName
        varchar FirstName
        varchar Title
        int ReportsTo FK
        datetime BirthDate
        datetime HireDate
        varchar Address
        varchar City
        varchar State
        varchar Country
        varchar PostalCode
        varchar Phone
        varchar Fax
        varchar Email
    }
    Genre {
        int GenreId PK
        varchar Name
    }
    Invoice {
        int InvoiceId PK
        int CustomerId FK
        datetime InvoiceDate
        varchar BillingAddress
        varchar BillingCity
        varchar BillingState
        varchar BillingCountry
        varchar BillingPostalCode
        decimal Total
    }
    InvoiceLine {
        int InvoiceLineId PK
        int InvoiceId FK
        int TrackId FK
        decimal UnitPrice
        int Quantity
    }
    MediaType {
        int MediaTypeId PK
        varchar Name
    }
    Playlist {
        int PlaylistId PK
        varchar Name
    }
    PlaylistTrack {
        int PlaylistId PK
        int TrackId PK
    }
    Track {
        int TrackId PK
        varchar Name
        int AlbumId FK
        int MediaTypeId FK
        int GenreId FK
        varchar Composer
        int Milliseconds
        int Bytes
        decimal UnitPrice
    }

    %% Relationships
    Album ||--o{ Artist : "belongs to"
    Track ||--o{ Album : "belongs to"
    Track ||--o{ MediaType : "uses"
    Track ||--o{ Genre : "categorized as"
    PlaylistTrack ||--o{ Playlist : "contains"
    PlaylistTrack ||--o{ Track : "contains"
    Invoice ||--o{ Customer : "issued to"
    InvoiceLine ||--o{ Invoice : "part of"
    InvoiceLine ||--o{ Track : "tracks"
    Customer ||--o{ Employee : "assisted by"
    Employee ||--o{ Employee : "reports to"

LangChain Setup to Chat with DB

Make a venv, install the requirements and load the API keys:

python3 -m venv datachat_venv #create the venv Linux
#python -m venv datachat_venv #create the venv W

#datachat_venv\Scripts\activate #activate venv (windows)
source datachat_venv/bin/activate #(linux)

pip install -r requirements.txt

#source .env

#export OPENAI_API_KEY="your-api-key-here"
#set OPENAI_API_KEY=your-api-key-here
#$env:OPENAI_API_KEY="your-api-key-here"

#echo $OPENAI_API_KEY

Wrapping Up LangChain with DB

I could replicate the working code from AlejandroAO by using:

1. The v1.4.5 sample DB version.
2. Python 3.12.3 for x86 and venvs
3. Im using MySQL 8.0

We can put all of this into containers with this docker-compose:

###docker exec -it mysql_db bash
###docker exec -it mysql_db mysql -u myuser -p chinook

FAQ

Other Tools to Chat over custom data (and locally~) is PrivateGPT

You can also try:

• PandasAI which I covered

• LlamaIndex

• https://console.anthropic.com/settings/keys

  • https://docs.anthropic.com/en/docs/about-claude/models

• https://platform.openai.com/api-keys

What it is LangGraph?

• https://github.com/langchain-ai/langgraph

Build resilient language agents as graphs - MIT!

LangGraph is a framework designed for building multi-agent systems, particularly when using large language models (LLMs).

It allows you to create complex workflows by defining tasks as nodes in a directed graph, with edges representing the flow of data.

• The system supports dynamic state management, maintaining context and ensuring coherent interactions across different agents.
• Fault tolerance is built in, meaning the system can recover from failures in individual agents.
• LangGraph is flexible and customizable, supporting custom agent logic and communication protocols.
• It is scalable, suitable for enterprise-level applications, and excels in handling complex workflows with a high volume of interactions.
• The framework is particularly useful for applications like chatbots, retrieval-augmented generation (RAG) systems, and self-correcting agents.