LangGraph Expense tracker

In short

Small project exploring the possibilitys of LangGraph. It lets you sent pictures of invoices, it structures and categorize the expenses and puts them in a database.

Visual overview of the project

To zoom visit the whiteboard here.

Some invoice extraction in action

Project structure

.
├── LICENCE.txt
├── README.MD
├── config.yaml
├── data
│   └── walmart-bon.jpeg
├── documents
│   ├── api_documentation.png
│   ├── langgraph.png
│   ├── openapi.json
│   └── pg_admin_screenshot.png
├── env.example
├── requirements.txt
└── src
    ├── api
    │   ├── __init__.py
    │   ├── category_routes.py
    │   ├── expenses_routes.py
    │   ├── payment_methods_routes.py
    │   └── run_api.py
    ├── chain
    │   ├── __init__.py
    │   ├── graphstate.py
    │   ├── helpers
    │   │   ├── __init__.py
    │   │   └── get_payment_methods_and_categories.py
    │   └── nodes
    │       ├── __init__.py
    │       ├── categorizer.py
    │       ├── correct.py
    │       ├── db_entry.py
    │       ├── humancheck.py
    │       ├── imageencoder.py
    │       └── jsonparser.py
    └── database
        ├── __init__.py
        ├── create_categories_and_payment_methods.py
        ├── create_tables.py
        └── db_connection.py

9 directories, 30 files

Step by step

1. Create Project

1.1 Create virtual environment
Using Conda, venv or any other tool of your liking.

1.2 activate virtual environment

1.3 clone repo
!TO DO!

1.4 install requirements
!TO DO!

1.5 create .env file
See example here.

---

2. Set up the database

2.1 Prerequisites

2.1.1 Install postgresql:

brew install postgresql

(other ways to install Postgresql)

2.1.2 Install Docker:

brew install docker

(other ways to install Docker)

2.2 Make a Docker container

2.2.1 Create:

docker run -d \
  --name postgres-expenses \
  -e POSTGRES_USER=expenses \
  -e POSTGRES_PASSWORD=money$ \
  -e POSTGRES_DB=expenses \
  -p 6025:5432 \
  postgres:latest

2.2.2 Control:
Use the following command to see if the container is running correctly:

docker ps

it should show a list of running containers.

2.3 Configure database

2.3.1 Create tables
Add tables for our expense tracking by running the /src/database/create_tables.py script (link)

2.3.2 Inspect tables
Using a tool link PGAdmin, you can inspect if the tables in the database are all there.

3. set up API

-Go to the root folder of your project and activate virtual environment

CD path/to/your/projectfolder
workon expense-tracker

i have some shell aliases set up, the [workon] command should probably be something like [conda activate] or [source [env]]

-activate virtual environment

(expense_tracker)
~/Developer/expense_tracker
▶ uvicorn src.api.run_api:app --reload
INFO:     Will watch for changes in these directories: ['/Users/jw/developer/expense_tracker']
INFO:     Uvicorn running on http://127.0.0.1:8000 (Press CTRL+C to quit)
INFO:     Started reloader process [12588] using StatReload
INFO:     Started server process [12590]
INFO:     Waiting for application startup.
INFO:     Application startup complete.

You can visit http://localhost:8000/docs#/ for a page with documentation about the API:

Database

The database consists of three main tables: categories, payment_methods, and expenses.

Table: categories

This table contains a list of categories for expenses. Each category has a unique ID and a name.

• Columns:
  • category_id (SERIAL, Primary Key): The unique ID for the category.
  • category_name (VARCHAR(100), Unique): The name of the category.

Table: payment_methods

This table contains various payment methods that can be used for expenses.

• Columns:
  • payment_method_id (SERIAL, Primary Key): The unique ID for the payment method.
  • payment_method_name (VARCHAR(50), Unique): The name of the payment method.

Table: expenses

This is the main table for tracking expenses. It contains information such as the date, the category (with a reference to the categories table), the payment method (with a reference to the payment_methods table), the amount, VAT, and other details.

• Columns:
  • transaction_id (SERIAL, Primary Key): The unique ID for the transaction.
  • date (DATE): The date of the expense.
  • category_id (INTEGER, Foreign Key): Reference to the categories table.
  • description (TEXT): A short description of the expense.
  • amount (DECIMAL(10, 2)): The amount of the expense.
  • vat (DECIMAL(10, 2)): The VAT for the expense.
  • payment_method_id (INTEGER, Foreign Key): Reference to the payment_methods table.
  • business_personal (VARCHAR(50)): Indicates whether the expense is business or personal.
  • declared_on (DATE): The date when the expense was declared.

API

See API documentation here: openapi.json

LangChain

the way this chain works is best described by showing the LangSmith Trace: click here to have a look