AI Triage System

Inspiration

Summer 2025 , I visited my village in Ballia district of Uttar Pradesh , India . I was taking a stroll ,the temperature was around 38 degree celcius , I saw a Public healthcare centre (PHC -a small rural government hospital) , ouside the centre there was a long queue of villagers standing in the heat, to get themselves checked by the doctor. What I saw that there were few patients whose condition was really critical but they had to wait for there chance to get checked buy the only doctor present at the hospital. Eventually their chance came and they got the treatment . This is where I got the idea of building something simple yet powerful that could help reduce this wait time and treat the critical patients at the earliest with an AI Powered Triage System. This system is based on a dataset which is summer specific for rural regions , with Symptoms people face during summers.

What it does

PHC's in India have max 3-5 workers at a centre with one doctor , one nurse and other staff maybe another ASHA worker from the village. This project help to fill this gap of less support staff but not more waiting time for critical patients.

The AI Triage System takes patient details and vitals (like heart rate, blood pressure, SpO₂, temperature, and symptoms), runs them through an AI model ,basically a Deep Learning Pipeline , and instantly classifies the patient into Green (Normal), Yellow (Urgent), or Red (Critical). So the patient who is the most critical is treated first. A healthcare worker has the access to this application , they go to the patients standing in the queue while the doctor is treating another patient , and feed their vitals to the system.

• The triage result is shown clearly to the healthcare worker.
• Data is automatically sent to the doctor, saving time.
• If a patient is Yellow or Red, an SMS alert is sent to the registered family member so they can immediately come and check on the patient.
• The doctor dashboard highlights urgent/critical cases so they can prioritize better.
• The system is bilingual (English + Hindi), mobile-friendly, and designed for rural usability.

How I built it

I Build the Backend using Python and FastAPI for APIs, with DL model for triage prediction. For Frontend I used streamlit , it gives a basic minimal frontend to test and also for prototype usage. Did SMS Integration Using TextBee API to send Automaticalerts to family members when a patient is critical. Also Doctor Dashboard Displays patients in real-time with triage filters, highlighting urgent cases.

Challenges I ran into

The first and the major challenge was DATA I visited several such helath centers to collect the patient data , but didn't get anything , at most of the centers there were no medical history of people were maintained , while at some the data was all unstructured and useless for training a AI model.Data is the main focus for any such project , without it nothing would come together properly. Secondly making the system usable by rural healthcare workers with limited digital literacy.

Accomplishments that I'm proud of

Succeded in making a **Doctor verified Dataset* - Initially I used a online available dataset just to conveniently scaffold a realistic shaped table , later on with the assistance of a Doctor tweaked the data, set correct vital limits for temperature , breathing rate , etc.

Succeded in **testing this System at a healthcare center* - Tested it out on several people and got an triage lable accuracy of around 94% , reduing the wait time for sritical patients by ~70-80 %.

• Built a working end-to-end AI triage system that connects patients, doctors, and families.
• Added an SMS alert system for families — bridging a real gap in rural healthcare.
• Created a minimal bilingual UI that is accessible and friendly for rural workers.
• Showcased how AI can directly improve healthcare delivery in underserved areas.

What I learned

I learned how a simple obesrvation and focusing on it can solve a big problem that goes unnoticed , that helped in not just saving time but saving time to save lives. Along with that I learned Importance of user experience design for rural communities (simplicity matters more than fancy features).

• Practical challenges of integrating SMS providers in India.

What's next for AI Triage System

• Integration with ABHA IDs (Aarogya Health IDs) to connect with India’s National Digital Health Mission.
• Adding regional language support (Tamil, Bengali, Marathi, etc.) beyond Hindi/English..
• Adding voice input/output for healthcare workers and patients with low literacy.
• Expanding dataset with more patient records to further improve triage accuracy.

Built With

• fastapi
• keras
• python
• streamlit
• tensorflow