Randchain

Randchain is a Python library for building intelligent decision trees and sequential processing chains powered by Large Language Models (LLMs). It simplifies the creation of automated workflows that can make decisions and execute actions based on natural language inputs.

Features

• Decision Tree Classification: Create dynamic decision trees that use LLMs to classify inputs and follow predefined paths
• Sequential Processing: Build chains of processing layers that transform inputs through multiple steps
• LLM Integration: Support for Ollama and OpenAI models
• Output Actions: Built-in support for sending emails, WhatsApp messages, and writing to files
• Modular Design: Clean separation into decision_tree and sequential modules

Installation

1. Clone the repository:

git clone https://github.com/pydevpk/randchain.git
cd randchain

2. Make sure you have Python>=3.10 and created environment for it.

3. Set up environment variables in a .env file:

OPENAI_API_KEY=your_openai_api_key_here
sender_email=your_email@example.com
password=your_app_password
account_sid=your_twilio_account_sid
auth_token=your_twilio_auth_token
from_number=whatsapp:+1234567890
to_number=whatsapp:+0987654321

Modules

decision_tree

The decision_tree module provides classes for building LLM-powered decision trees.

Key Classes:

• LLM: Wrapper for LLM interactions with Ollama or OpenAI
• DTCliassifier: A node in the decision tree that makes decisions using LLMs

Usage:

from decision_tree import DTCliassifier, LLM

# Create a decision node
decision_node = DTCliassifier(
    "MyAgent",
    LLM("Ollama", "qwen2.5:latest", "Your prompt here")
)

# Make a decision
result = decision_node.decide("User query here")

sequential

The sequential module allows you to create chains of processing layers.

Key Classes:

• Layer: Represents a single processing step with configurable inputs and outputs
• Sequential: Manages a sequence of layers that process inputs progressively

Usage:

from sequential import Layer, Sequential

chain = Sequential([
    Layer(my_function, whatsapp_config=whatsapp_config),
    Layer(another_function)
])

result = chain.run(["input1", "input2"])

Note: First layer takes input from provided list of inputs, then next layers will take input from previous layer's output.

Configuration

Use the config.py classes to set up communication channels:

from config import EmailConfig, Whatsappconfig

email_config = EmailConfig(
    sender_email="sender@example.com",
    password="password",
    recipient_email="recipient@example.com",
    subject="Automated Message"
)

whatsapp_config = Whatsappconfig(
    account_sid="your_account_sid",
    auth_token="your_auth_token",
    from_number="whatsapp:+1234567890",
    to_number="whatsapp:+0987654321"
)

Utilities

The utils module provides helper functions:

• send_email(): Send emails using SMTP
• send_whatsapp(): Send WhatsApp messages via Twilio
• txt_writer(): Write text to files
• output_conversion(): Convert outputs to desired types

Testing

Run the test suite:

python test.py

The test demonstrates an insurance eligibility system using a decision tree combined with sequential processing for WhatsApp notifications.

Dependencies

• ollama
• openai
• twilio
• python-dotenv

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License.