Omnilex Agentic Retrieval Competition Starter Repo

Official starter repo for Kaggle competiton https://www.kaggle.com/competitions/llm-agentic-legal-information-retrieval/host/launch-checklist

Quick Start

Installation

(Tested with Ubuntu-24.04 in WSL)

# Clone the repository
git clone https://github.com/Omnilex-AI/Omnilex-Agentic-Retrieval-Competition.git
cd Omnilex-Agentic-Retrieval-Competition

# Create and activate virtual environment
python3 -m venv .venv
source .venv/bin/activate  # Linux/macOS
# .venv\Scripts\activate   # Windows

# Install dependencies
pip install -r requirements.txt
pip install -r requirements-dev.txt  # for testing/linting

# Install package in development mode
pip install -e .

Download Data

Get it from Kaggle into data directory

Run Baselines

Two baseline notebooks are provided:

1. Direct Generation (notebooks/01_direct_generation_baseline.ipynb)

   • Prompts LLM to directly generate citations
   • Simple but prone to hallucination

2. Agentic Retrieval (notebooks/02_agentic_retrieval_baseline.ipynb)

   • Uses ReAct-style agent with search tools
   • Grounded in actual legal documents

Both notebooks work in VSCode and can be submitted to Kaggle.

Validate Submission

python scripts/validate_submission.py submission.csv

Data Format

See Kaggle

Project Structure

├── src/omnilex/           # Core library
│   ├── citations/         # Citation parsing & normalization
│   ├── evaluation/        # Metrics & scoring
│   ├── retrieval/         # BM25 search & tools
│   └── llm/               # LLM loading & prompts
├── notebooks/             # Baseline notebooks
├── utils/                 # Data & utility scripts
├── tests/                 # Test suite
└── data/                  # Data directory

Requirements

• Python >= 3.10
• llama-cpp-python (for local LLM inference)
• rank-bm25 (for keyword search)
• pandas, numpy, scikit-learn

For Kaggle submissions, you may need to (depending on your solution):

1. Upload your GGUF model as a Kaggle dataset
2. Upload pre-built indices as a Kaggle dataset
3. Package the omnilex library

License

Apache 2.0 - See LICENSE

Contact

For public questions about the competition please use the "Discussion" tab or open an issue on this repository. For private questions reahc out to host on Kaggle or ari.jordan@omnilex.ai