Saarthi

• 

  An example of the input

💡 Inspiration

One day, I was searching my symptoms online after having a headache and blurred vision. Within minutes, Google went from suggesting lack of sleep and dehydration to showing serious illnesses like cancer. 😵‍💫

The experience was overwhelming and anxiety-inducing. I realized that while there is plenty of health information online, there isn't always a simple way for people to describe their symptoms and receive structured, understandable insights.

That's when I decided to build Saarthi 🚑 — an AI-powered health companion that helps users understand their symptoms in a more organized and explainable way.

🩺 What it does

Saarthi allows users to enter symptoms in natural language, just like they would describe them to a friend or a doctor.

✨ Features:

Understands symptom descriptions using Medical NLP. Predicts the most likely diseases and conditions. Shows the Top-3 predictions with confidence scores. Generates easy-to-understand explanations using AI. Encourages users to seek professional medical advice when necessary. 🛠️ How I built it

I built Saarthi using a hybrid AI architecture:

🧠 BioClinicalBERT for medical language understanding 📊 Logistic Regression for disease prediction ⚡ Flask for the backend API 🤖 Groq + Llama for explainable AI responses 💻 HTML, CSS, JavaScript/TypeScript for the frontend interface 🐍 Python, Transformers, PyTorch, Scikit-learn for model development

Workflow User Symptoms ↓ BioClinicalBERT ↓ Disease Prediction ↓ Top Predictions ↓ AI Explanation Layer ↓ Health Insights 🚧 Challenges I ran into

🔹 Finding and working with medical datasets that were limited and sometimes noisy.

🔹 Making predictions accurate while keeping response times fast.

🔹 Integrating traditional machine learning with modern LLMs.

🔹 Handling vague symptom descriptions that could point to multiple conditions.

🔹 Designing the system to provide helpful insights without claiming to be a medical diagnosis tool.

🏆 Accomplishments that I'm proud of

✅ Built a complete end-to-end AI healthcare application as a solo developer.

✅ Successfully integrated BioClinicalBERT for medical NLP.

✅ Added an explainable AI layer instead of showing only raw predictions.

✅ Developed a full backend API and connected it with an interactive frontend.

✅ Created a scalable architecture that can be improved with larger medical datasets and advanced models.

📚 What I learned

🌱 How domain-specific NLP models like BioClinicalBERT differ from general-purpose language models.

🌱 The importance of data quality in machine learning projects.

🌱 How to combine machine learning models with LLMs to create more useful user experiences.

🌱 Building production-style AI pipelines with APIs, inference systems, and frontend integration.

🌱 The importance of responsible AI in healthcare-related applications.

🚀 What's next for Saarthi

🔮 Improve prediction accuracy with larger and more diverse medical datasets.

🩻 Add symptom normalization and medical knowledge graphs.

🎙️ Enable voice-based symptom input for accessibility.

📱 Create a mobile-friendly experience.

🌐 Deploy Saarthi publicly so anyone can access AI-powered symptom insights from anywhere.

🧑‍⚕️ Add doctor recommendations and healthcare resource integration in the future

Built With

• 3
• api
• bioclinicalbert
• css3
• flask
• git
• groq
• html5
• javascript
• joblib
• llama
• python
• pytorch
• rest
• scikit-learn
• transformers
• typescript