YUVA | Devpost

Inspiration

One of our team member's grandfather had a stroke, and they had a hard time communicating with their loved ones and caretakers. Without the proper resources, the ability to communicate is hindered.
We wanted to give everyone who has been through such a traumatic incident a chance to have a voice and be able to COMMUNICATE with people and express their emotions.

The app offers a visual, tap-based communication board, natural text-to-speech output, interactive speech therapy modules, and AI phrase prediction. It enables users to express their needs, practice speech, and rebuild confidence through our simple and accessible interface.
With features like real-time speech feedback, personalized phrase suggestions, cognitive-language therapy, and caregiver/therapist support, NeuroSpeak empowers users to reconnect with their loved ones, participate in conversations, and take control of their recovery.

We used React to build the frontend. For the backend, we used FastAPI+Uvicorn through Python, Google Cloud API platforms, and FireStore for the database. To connect frontend to backend we used Axios which is a library to get Fast API requests to React Native.

The challenge we ran into was connecting frontend to backend.
We initially wanted to add a "sign in with Google" option, but unfortunately, it was difficult to incorporate without changing the original sign-in or sign-up options, so we decided to come back to that feature if we had time after we completed everything else.
We accidentally git-committed a file that we were supposed to git-ignore which made a private file public, and we had to delete our current working service account, create a new one, and generate a new API key to call the Google Cloud platforms.

We are really proud that we were able to complete 90% of the project in the first 12 hours of the hackathon.

We learned how to create a full-stack AI-assisted application, combining a React frontend with a FastAPI backend, and integrating with Google Cloud Firestore for real-time data storage.
We implemented OAuth 2.0 authentication, enabling secure login via email while managing user sessions and storing user profiles in Firestore.
We explored AI-based phrase prediction and text-to-speech (TTS) systems to enhance accessibility and learn how to handle speech generation and correction feedback.
We developed feedback, allowing users to confirm or reject speech outputs, with reprocessing logic in the backend to improve accuracy over time.
We learned to debug and resolve issues like missing modules, misconfigured paths, or invalid Firestore writes — gaining deeper experience with dependency management and cloud services.