Inspiration
I lost a close family member to Diabetes suddenly, and in a way that felt painfully preventable. The early warning signs were there, but they were missed, buried in routine metrics no one had the time or tools to interpret. That loss shook me. It made me realize how many lives could change if clinicians had a simple system that could surface those red flags early especially during the pre-diabetic stage when intervention makes the biggest difference. StethX is my answer to that moment, a way to turn grief into a tool that might save someone else’s story from ending the same way.
What it does
StethX analyzes common patient metrics such as age, BMI, fasting glucose, HbA1c, and blood pressure using a trained machine learning model to predict early Diabetes risk. It then uses an LLM to:
- Explain why the patient is at risk
- Highlight the most influential factors
- Provide prevention-aligned recommendations
- Communicate in clear, clinician-ready language In short, it transforms raw numbers into real-time, actionable clinical insight.
How we built it
I trained a simple machine learning model on a Diabetes dataset to predict early risk based on metrics like BMI, glucose, and blood pressure. The prediction is sent to an LLM, which explains the reasoning in clear clinical language. The frontend was built with Next.js and TailwindCSS, and the backend exposes a lightweight API for both ML inference and LLM generation. Everything works together to deliver fast, understandable clinical insights in one clean dashboard.
Challenges i ran into
- Ensuring explanations were medically aligned without becoming overly complex.
- Needed to craft prompts that generated trustworthy, structured outputs.
- Real-world clinical data is sensitive so i had to choose interpretable models that worked well with available datasets.
- Cutting out unnecessary features to focus on a powerful, functional MVP.
Accomplishments that i am proud of
- Achieved reliable ML prediction accuracy of 96%
- Designed a clean, fast UI that feels like a real clinical dashboard
- Integrated an LLM to generate clinician-ready reasoning and suggestions
- Created a system that could genuinely support early intervention and save lives
What i learned
- How to combine ML predictions with LLM reasoning in a controlled, medically aligned way
- The importance of explainability in healthcare AI
- How to scope and execute a full clinical DSS under time pressure
- How UX can dramatically impact trust in medical tools
- That AI shines brightest when augmenting and not replacing human expertise
What's next for StethX
- Fine-tune an LLM on clinical notes for richer insights
- Partner with clinicians to refine recommendation quality
- Expand beyond Diabetes to other chronic condition early-warning systems