This repository, developed by Stefan Pietrusky, is based on the QUEST prototypes. The app lets users choose a topic and language, generates a spoken question via an LLM and Text-to-Speech, and plays it back. Users respond by uploading or recording audio, which is transcribed with Whisper, edited if needed, and then sent back to the server. The LLM analyzes the answer, returns written feedback (also converted to speech), and the user can reset the session to start over.
The application is currently being tested and further developed in studies.
⚠️ Work in Progress: This prototyp is currently under active development. While I make it available for research purposes, please be aware that there will be some changes to the functional structure. I recognize that some current technical design decisions may not be optimal and are subject to revision. Researchers using this prototyp should expect potential updates and changes. I recommend checking back regularly for updates and versioning information.
The structure of the current [V1] QUEST app is shown below.
Below is a short GIF showing the structure and function of the app.
In this version, the QUEST app's feedback is based on the specifications of the Common European Framework of Reference for Languages (CERF). Depending on the application, this can be adapted in the framework's code.
- Clone this repository on your local computer:
git clone https://github.com/stefanpietrusky/QUEST.git- Install the required dependencies:
pip install -r requirements.txt- Install Ollama and load the default model Llama3.2 (3B). Alternatively, another model can be used. In this case, the reference in the code must be adjusted.
ollama pull llama3.2python app.py