Adaptive Palette

This project builds a palette that empowers AAC(Augmentative and Alternative Communication) users with the ability to personalize it according to their specific requirements, thereby enhancing their communication capabilities with others.

The front end of the project is built with Preact.

Install

To work on the project, you need to install NodeJS and NPM for your operating system.

Then, clone the project from GitHub. Create a fork with your GitHub account, then enter the following in your command line (make sure to replace your-username with your username):

git clone https://github.com/your-username/adaptive-palette

From the root of the cloned project, enter the following in your command line to install dependencies:

npm ci

Development

Start a Server

To start a local web server, run:

npm start

To start a local web server for development that every change to the source code will be watched and redeployed, run:

npm run dev

The website will be available at http://localhost:3000.

Enable RAG

RAG (Retrieval-Augmented Generation) is an AI technique designed to enhance the accuracy of generative models by incorporating factual knowledge from external sources. It requires loading factual knowledge into a vector store that will be quried to provide relevant information to the language model as a context.

By default, the use of RAG is turned off in the system. The enableRag flag is set to false by default in the config/config.ts.

Follow these steps to complete a one-time setup to enable RAG in the system:

1. Load a document into the vector store

   Use the scripts/loadDocIntoVectorDb.js script to populate the vector store. Run the following command from the project root directory:

   node scripts/loadDocIntoVectorDb.js [location-of-document] [target-dir-of-vector-db]

2. Configure the application

   Update the config/config.ts file to specify the path to the vector store directory and set the flag enableRag to true:

   export const config = {
     // ... other configurations
     rag: {
      enableRag: true,
       vectorStoreDir: "[path-to-vector-db-directory]"
     }
     ...
   };

   Note: The vectorStoreDir defaults to ./vectorStore. Modify the value to match where your vector store is located. When a relative path is used, the path is relative to the project root directory.

3. Restart the server

   Follow the instruction in the Start a Server section.

Lint

To lint the source code, run:

npm run lint

Run Tests

To run tests, run:

npm test

Demonstrations

The sub-folder demos contains code for a number of demonstrations. These are short examples. The apps folder contains more fully built-out application examples. See the respective READMEs for instructions on how to run the software.

• Ollama Chat Web-App: a chat application running on localhost that provides access to multiple LLMs using the Ollama localhost web service.
• Palette Generator Web-App: an application for generating and saving a palette using the Bliss gloss. By providing a set of gloss words, BCI AV IDs, or svg builder strings, the Bliss gloss is searched and a palette is generated based on matches found.
• Ollama Chat Service Demo: a simple web-app that runs on localhost for sending queries to an Ollama chatbot service also running on localhost.

Cloudflare

The adaptive-palette can be served as a production preview using Cloudflare Pages, specifically using the Git integration guide. You will need to have your own Cloudflare account to do this.

In the "Deployment details" for the preview, use the following for the "Build command" and "Build output directory" settings:

• Build command: npm run build:client
• Build output directory:: /dist/client

Utility Scripts

• Load a document into a vector databbase (scripts/loadDocIntoVectorDb.js)
• Retrieve and store Bliss-character status and composition information (scripts/tsvFindCompositions.js)