txn-harvester

Extract and structure financial transaction data from unstructured text

txn_harvester is a Python package designed to parse and validate financial transaction data from raw text inputs (e.g., bank statements, transaction logs) into structured formats. It leverages LLM7 (via langchain_llm7) by default, but supports any LangChain-compatible LLM for flexibility.

🚀 Features

• Extracts transaction details (amount, date, description, category) from unstructured text
• Validates output against predefined financial patterns using regex
• Supports custom LLMs (OpenAI, Anthropic, Google, etc.) via LangChain
• Lightweight and easy to integrate into financial workflows

📦 Installation

pip install txn_harvester

🔧 Usage

Basic Usage (Default LLM7)

from txn_harvester import txn_harvester

user_input = """
Paid for groceries at Whole Foods: $125.50 on 2024-05-15
Rent payment: $1500.00 (due 2024-05-20)
"""

response = txn_harvester(user_input)
print(response)

Custom LLM (OpenAI Example)

from langchain_openai import ChatOpenAI
from txn_harvester import txn_harvester

llm = ChatOpenAI(model="gpt-4")
response = txn_harvester(user_input, llm=llm)

Custom LLM (Anthropic Example)

from langchain_anthropic import ChatAnthropic
from txn_harvester import txn_harvester

llm = ChatAnthropic(model="claude-2")
response = txn_harvester(user_input, llm=llm)

Custom LLM (Google Example)

from langchain_google_genai import ChatGoogleGenerativeAI
from txn_harvester import txn_harvester

llm = ChatGoogleGenerativeAI(model="gemini-pro")
response = txn_harvester(user_input, llm=llm)

🔑 API Key Configuration

• Default: Uses LLM7_API_KEY from environment variables.
• Manual: Pass via api_key parameter or set LLM7_API_KEY in your shell:

  export LLM7_API_KEY="your_api_key_here"
  

• Free Tier: Sufficient for most use cases (rate limits apply).
• Get Key: Register at https://token.llm7.io

📝 Parameters

📝 Output

Returns a list of structured transaction data (e.g., [{"amount": "$125.50", "date": "2024-05-15", ...}]).

🔄 Customization

• Modify regex patterns in prompts.py to adapt to specific transaction formats.
• Extend the package by subclassing txn_harvester for domain-specific parsing.

📝 License

MIT

📧 Support & Issues

• GitHub Issues: https://github.com/chigwell/txn-harvester/issues
• Author: Eugene Evstafev (hi@euegne.plus)