DoctorAI-QA

Inspiration

Millions of people search for health information online but often encounter misinformation or confusing advice. We wanted to create an open-source AI tool that provides clear, reasoning-based, and educational answers to healthcare questions, making reliable medical knowledge accessible to students, educators, and the public.

What it does

DoctorAI-QA is a GPT-OSS model fine-tuned on healthcare datasets using LoRA and Unsloth. It answers questions about diseases, symptoms, medications, lifestyle, and more in a human-readable, educational format. Users can interact with the model through a Gradio or Colab demo, making it easy to explore healthcare topics safely for learning purposes.

How we built it

• Base model: GPT-OSS 20B (unsloth/gpt-oss-20b-unsloth-bnb-4bit)
  
• Fine-tuning: LoRA adapters on medical datasets for fast reasoning
  
• Interface: Gradio for interactive testing; also runnable on Colab
  
• Hosting & sharing: Model and weights on Hugging Face, code on GitHub
  
• License: Apache-2.0 (fully open-source)

Challenges we ran into

• Ensuring responses were accurate and safe for educational purposes
  
• Filtering and formatting healthcare datasets for quality fine-tuning
  
• Keeping the model lightweight and fast while maintaining reasoning ability
  
• Presenting answers in a clear, user-friendly interface

Accomplishments that we're proud of

• Built a fully open-source healthcare QA AI with real-time interactive demos
  
• Optimized the model using LoRA + Unsloth for efficiency and speed
  
• Created a tool that can be used by students and educators worldwide
  
• Successfully demonstrated a working MVP accessible via Gradio and Colab

What we learned

• Fine-tuning large language models requires careful dataset preparation and evaluation
  
• Clear user interface and educational framing are critical for accessibility
  
• Open-source collaboration accelerates development and reproducibility
  
• AI can provide useful educational insights when responsibly applied to healthcare

What's next for DoctorAI-QA

• Add multilingual support for non-English users
  
• Expand the dataset to cover mental health and preventive care
  
• Integrate with a lightweight chatbot interface for easier interaction
  
• Gather feedback from educators and students to improve usability and accuracy

Built With

• ai
• collab
• gpt
• gptoss
• gradio
• huggingface
• llm
• lora
• modelai
• openai
• unsloth