hapax 0.1.0

A flexible workflow system for LLM applications

Hapax

A flexible workflow system for LLM applications with built-in observability.

Features

• Simple, intuitive workflow creation
• Built-in OpenTelemetry tracing
• Type-safe workflow validation
• Parallel and conditional execution
• LangChain integration

Installation

pip install hapax

Quick Start

from hapax.workflow import Workflow

# Create a simple workflow
workflow = (
    Workflow()
    .add(load_document)
    .add(split_text)
    .add(create_embeddings)
    .add(search_relevant)
    .add(generate_answer)
)

# Run with tracing
result = await workflow.run(
    input={"document": doc, "question": "What is...?"},
    trace=True
)

RAG with Observability Example

The examples/rag_with_observability.py demonstrates how to build a Retrieval-Augmented Generation (RAG) pipeline with built-in observability using Hapax. This example showcases:

• Processing both text and tabular data
• Using LangChain's MultiVectorRetriever for improved retrieval
• Adding observability using the @obs decorator
• Error handling and state management
• Type-safe workflow validation

To run the example:

1. Install required dependencies:

pip install -r requirements.txt

2. Set your OpenAI API key:

export OPENAI_API_KEY=your-api-key

3. Run the test script:

python examples/test_rag_pipeline.py

The example will demonstrate:

• Loading and processing semi-structured data (text and tables)
• Generating summaries for improved retrieval
• Answering questions using both context types
• Observability metrics and tracing throughout the pipeline

Development

1. Clone the repository:

git clone https://github.com/teilomillet/hapax-py.git
cd hapax-py

2. Create a virtual environment and install dependencies:

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev,test]"

3. Run tests:

pytest tests/ -v

License

MIT License