Inspiration
Healthcare information is often scattered, complex, and not easily understandable for everyone. We wanted to build an AI assistant that provides quick, reliable, and easy-to-understand answers to common medical questions. The goal was to empower patients, students, and curious minds with a safe educational tool.
What it does
DoctorAI-QA is a fine-tuned GPT-OSS model specialized in healthcare Q&A. It provides reasoning-based, conversational answers to medical queries such as symptoms, conditions, and lifestyle advice. The model runs interactively through Colab or Gradio and is designed for educational and demonstration purposes only.
How we built it
- Started with GPT-OSS base model.
- Fine-tuned using LoRA adapters for efficiency.
- Leveraged Unsloth for 2x faster training.
- Integrated a healthcare dataset with disease, symptoms, and treatment questions.
- Built a Colab + Gradio interface for easy testing and demo.
- Deployed model weights to Hugging Face for public access.
Challenges we ran into
- Limited GPU runtime on Google Colab, which caused training interruptions.
- Managing large model weights and pushing them to Hugging Face repositories.
- Ensuring answers were safe, non-prescriptive, and framed for educational purposes only.
- Debugging runtime errors caused by GPU limits when re-running the notebook.
Accomplishments that we're proud of
- Successfully fine-tuned a large GPT-OSS model on healthcare data.
- Created a clean Gradio demo for real-time interaction.
- Open-sourced both the Colab notebook and model weights for others to explore.
- Learned to optimize large model training using Unsloth + LoRA within limited resources.
What we learned
- Hands-on experience in fine-tuning LLMs with LoRA.
- How to integrate models into interactive demos with Gradio and Hugging Face.
- The importance of balancing technical performance with usability and safety.
- Improved understanding of GPU memory constraints and efficiency hacks.
What's next for DoctorAI-QA
- Expand the dataset to cover more diverse healthcare topics.
- Add multi-language support for accessibility.
- Build a lightweight version deployable on mobile or low-resource environments.
- Improve the demo with visual dashboards and better UI/UX.
- Explore collaboration with medical educators to refine educational value.
Log in or sign up for Devpost to join the conversation.