ECLAIR

Enterprise sCaLe AI for woRkflows

[email protected]

Example of ECLAIR running on a real-world nursing workflow in Epic after being given only a video recording and natural language description of the task. Note that this is sped up from actual model execution.

🤖 Multimodal foundation models (FMs) such as GPT-4 offer a promising approach for end-to-end workflow automation given their generalized reasoning and planning abilities.

⚙️ To study these capabilities we propose ECLAIR, a system to automate enterprise workflows with minimal human supervision.

📊 Our initial experiments suggest that ECLAIR can overcome the limitations of traditional automation technologies (e.g. RPA) with (1) near-human-level understanding of workflows and (2) instant set-up with minimal technical barrier.

Please note that ECLAIR is an ongoing research project and is not production-ready.

💿 Installation

# Create virtual env
conda create -n eclair_env python=3.10 -y
conda activate eclair_env

# Install repo
git clone https://github.com/HazyResearch/eclair-agents.git
cd eclair-agents/
pip install -r requirements.txt
pip install -e .

📊 Paper Experiments

Generate the experimental results in our paper using the dataset + scripts in this section.

💾 Data

• Link to Data -- Download this file into the data/ folder and unzip it.
• You should now have a folder at data/vldb_experiments.

🚀 How to Run

export OPENAI_API_KEY=<your_openai_api_key>

# Demonstrate
bash eclair/vldb_experiments/demonstrate_experiments/run_experiments.sh

# Execute
bash eclair/vldb_experiments/execute_actions/run_experiments.sh
bash eclair/vldb_experiments/execute_grounding/run_experiments.sh # [TODO]

# Validate
bash eclair/vldb_experiments/validate_experiments/run_experiments.sh

🏥 Hospital Workflow

This section contains the workflow data and scripts used to automate a real-world nursing workflow in Epic (i.e. the demo video at top of this README).

🎥 Demo

• Link to ECLAIR Demo -- Visit this folder to view the outputs of ECLAIR executing the nursing workflow in Epic.
• Please note that there are two versions of the demo video -- the raw recording as well as a 10x sped up version (labeled as [fast]).

💾 Data

• Link to Data -- Download this folder into the data/ folder.
• You should now have a folder at data/hospital_data.

🚀 How to Run

This will run the end-to-end automation pipeline for the nursing workflow.

First, it generates an SOP from a demonstration. Second, it runs ECLAIR on the given workflow. Third, it validates that the workflow was completed successfully.

Note that this assumes you have a sandboxed instance of Epic running on your computer.

cd eclair/agent
python3 pipeline.py

Citation

Please consider citing if you found this work or code helpful!

@misc{wornow2024automating,
      title={Automating the Enterprise with Foundation Models}, 
      author={Michael Wornow and Avanika Narayan and Krista Opsahl-Ong and Quinn McIntyre and Nigam H. Shah and Christopher Re},
      year={2024},
      eprint={2405.03710},
      archivePrefix={arXiv},
      primaryClass={cs.SE}
}