Lethelink is an interface to create grounding messages for people who suffer from anterograde amnesia, which a popular symptom of Alzheimer's. The messages can be delivered to the person's hearing aid on a schedule to prevent episodes of disorientation.

Caregivers can use Lethelink to create nudges that ground their Patient in reality, remind them they are safe, and help them orient in time and space.

🌿 Anchors are context-aware. For example, they can learn what causes the Patient to feel anxious, or what their schedule looks like today.

🌿 With one voice sample, Lethelink can send anchors in the Caregiver's voice, which is more familiar to the Patient.

🌿 Prevent the onset of disorientation by selecting what causes an anchor to send: time, Patient's voice, or increased heartbeat.

The Caregiver can set the context for the anchor through a UI that looks like this:

In this demo, we show how a nudge generated from the voice of our teammate could help his elderly mother avoid the state of disorientation.

Note: Lethelink v.0.1 is an MVP we built in one day. Its main purpose is to demonstrate the possibilities of combining powerful AI models in a thoughtful interface. This is by no means a finished product, and we encourage others to reuse our code and build on it. We may continue to build this out in the future.

Lethelink is fully open-source, so you can build on it. Here are some ideas we are excited about:

Engineering

• Send nudges when the person's heartbeat spikes. Get data from their wearable.
• Build a more streamlined hearing aid integration.
• Create an end-to-end UI that connects all the pieces together.

Research

• Start an evidence-based clinical practice using nudges to help people with Alzheimer's.
• Research if people with amnesia learn to anticipate Lethelink and rely on it.

Note: We walk through every step in this guide, but if you don't code and are not sure what to do, reach out to Yaroslav to get help with the setup.

• An account at ElevenLabs
• An account at OpenAI

1. Environment Variables Setup Clone the Lethelink repository to your local machine. Inside the project directory, you need to set up your environment variables:

   • ELEVEN_LABS_API_KEY: API key generated from Eleven Labs.
   • OPENAI_API_KEY: API key generated from OpenAI.

   Use a .env file to store these variables and python-dotenv to load them.

2. Set Voice Preference In app.py, a default VOICE_ID for "Emily" is provided. This voice ID is specific to Eleven Labs. Optional: If you want to fine-tune on the Caregiver's voice, go to this page, add a clear voice sample, and train the voice. Then update the VOICE_ID variable with the new ID.

3. Install Dependencies Use the provided command to install the required dependencies:

   pip install openai streamlit mutagen requests python-dotenv

4. Run the Application With Streamlit, run the app using the following command:

   streamlit run app.py

   By default, the application will be accessible at http://localhost:8501.

   Because we expect most users to first experiment with promts and voices, Lethelink will help you save credits by caching generated text and audio to your local directory. So if you send the same prompt with the same voice, you will get an immediate, free response.

• Custom Prompts: Modify the generate_nudge function in app.py to customize the prompting mechanism.
• Voice Options: Check Eleven Labs' documentation for different voice options and update the VOICE_ID accordingly.
• Integrate with Other Models: The application is currently set up for OpenAI's GPT-4 model. If you have access to different models, adjust the call_chatgpt function in the utils module.

• License: MIT (feel free to use, modify, and distribute this code as you see fit).

Lethelink is fully open-source, and we welcome contributions! See Opportunities for ideas on what to contribute.