Brave Buddy - An AI Voice Companion for Pediatric Cancer Patients

A voice-based AI assistant designed to emotionally support paediatric cancer patients before surgery. Built with GPT-4o, Whisper AI, and NeMo Guardrails, Brave Buddy enables empathetic, safe, and personalised conversations — from answering medical questions in a child-friendly tone to reducing pre-op anxiety through emotionally intelligent dialogue and reminders.

The Problem:

Hospitals can be an overwhelming space, especially for children undergoing cancer treatment. Anxiety, confusion, and emotional stress are common—yet often under-addressed. Brave Buddy was created to bridge that gap with an AI-powered voice companion that speaks the child’s language, understands emotional cues, and ensures accurate, medically-safe conversations.

Our Solution:

We developed a working prototype powered by GPT-4o for emotionally supportive dialogue and Whisper AI for real-time speech processing. To make Brave Buddy truly fit for pediatric support, we fine-tuned GPT-4o on 500+ real-world counseling transcripts, teaching it to generate responses that were both emotionally supportive and age-appropriate. This fine-tuning enabled the assistant to better reflect the tone, empathy, and vocabulary needed for children facing serious health challenges.

We used NVIDIA’s NeMo Guardrails to enforce safety and ethical constraints, ensuring that all responses remained compliant and responsible. Function calling was used to support:

• Storing and recalling preferences for enhanced user customizability

• Creating reminders

• Tracking emotional sentiment over time

• Handling medically grounded Q&A through a RAG system tuned for clarity and empathy

To enhance interactivity, we reduced latency by 40% using multithreaded TTS and asynchronous response chunking— making the experience feel more natural and responsive.

Our dream team:

The project was developed by a diverse, cross-functional team of 11, including engineers, medical professionals, and business students. It earned 2nd place at the Rutgers Health Hackathon and received over 90% positive feedback from survey participants who saw potential in Brave Buddy as a supportive tool for pediatric care.

The story isn't over yet:

We plan to transition the prototype to AWS for HIPAA compliance, integrate services into a containerized architecture, and continue fine-tuning the assistant based on clinician input and real usage scenarios.

Tech Stack

• LLM: GPT-4o (fine-tuned on paediatric counselling data)

• Voice Stack: Whisper API (STT/TTS)

• Guardrails: NeMo SDK for ethical & safe conversations

• Function Calling: Reminders, personalization, check-ins, Q&A, sentiment scoring

• RAG System: For medical knowledge retrieval in child-friendly tone

• Optimizations: Multithreaded TTS, async response chunking

For more details, checkout our Github page: https://github.com/Seninfarheen/BraveBuddy