AI Psychiatrist Assistant: An LLM-based Multi-Agent System for Depression Assessment from Clinical Interviews

Environment Setup

1. Clone the repository:

git clone https://github.com/trendscenter/ai-psychiatrist.git

2. Navigate to the cloned directory and create a new conda environment using the provided env_reqs.yml file:

cd ai-psychiatrist
conda env create --name aipsy --file ./env_reqs.yml

3. Activate the conda environment:

conda activate aipsy

4. Create a new git branch for your changes:

git checkout -b dev_<your_last_name>

Replace <your_last_name> with your last name.

Ollama on TReNDS Cluster

1. Start Ollama by submitting the SLURM job script job_ollama.sh:

cd slurm
sbatch job_ollama.sh

2. Check the status of the job using the command squeue -u <username>, where <username> is your username. Check the node that Ollama is running on in the output of the command. The node name is in the format arctrdagnXXX, where XXX is a number.

3. Once the job is running, you can access Ollama on the node. See the Python script ollama_example.py for an example of how to use Ollama. Update OLLAMA_NODE to the node where Ollama is running. Submit the SLURM job script job_assess.sh to run the Python code:

sbatch job_assess.sh

4. If you stop using Ollama, you can stop the job using the command scancel <job_id>, where <job_id> is the job ID of the Ollama job. You can find the job ID in the output of the command squeue -u <username>.

Code

• qualitative_assessment contains the scripts for identifying social and biological risk factors.

• quantitative_assessment contains the scripts for predicting PHQ-8 scores.

• meta_review contains the code for integrating information and predicting severity.

• agents contains the code for the multi-agent system.

• analysis_output contains the outputs from the quantitative assessment.

• visualization contains the scripts for generating the figures.

References

• OpenReview paper

• Ollama documentation

• TReNDS cluster documentation