Telemedicine

AI-Powered Telemedicine Platform

Visit Project

Project Overview

A full-stack, intelligent healthcare platform designed to bridge the gap between patients and doctors, especially in low-connectivity environments. This project leverages a microservices architecture and advanced AI to provide accessible, real-time, and asynchronous medical consultations. [Patient: tester@example.com, 1234567890], [Doctor: testing@example.com, 1234567890]

Key Features I Implemented

Live & Asynchronous Consultations

Built a dual-mode system allowing for real-time video calls with doctors and AI powered auto report generation from the call and an asynchronous "MedReach" feature for patients to send video/audio messages when doctors are unavailable.

Offline-First Transcription

Integrated a client-side Speech-to-Text model (Whisper) that transcribes patient recordings directly in the browser, ensuring functionality even in poor internet conditions.

AI-Powered Emergency Triage

Developed an intelligent triage system that analyzes patient transcripts for severity. It automatically routes high-severity cases to doctors and mobile clinics, and medium-severity cases to community health workers (ASHA).

Emergency Fallback System

Engineered a critical safety net that alerts public sector units (police/fire) for transport or oxygen support if no medical staff are available to respond to an emergency.

Multilingual AI Chatbot

Created a RAG-based medical chatbot using Google Gemini for real-time translation and accurate responses. Also built a k-NN model to predict potential ailments from user-reported symptoms.

Unified Health Records

Designed a centralized system for patients to access and download their health records from both online consultations and physical hospital visits.

How I Solved Key Healthcare Challenges

Challenge: The Problem of Poor Connectivity

Traditional telemedicine fails in rural areas.

My Solution:

Built the system with an offline-first approach. By processing transcription on the client's device, the platform can send a lightweight text file instead of a large video file over a weak network, ensuring no patient is left behind.

Challenge: The Danger of Delayed Emergency Response

Patients often don't know how serious their condition is.

My Solution:

The AI triage system that instantly analyzes symptoms and flags emergencies, bypassing queues and routing alerts directly to the nearest available medical help, significantly cutting down response times.

Challenge: The Barrier of Language & Literacy

Patients couldn't communicate effectively.

My Solution:

The multilingual AI chatbot, allowing users to describe their symptoms in their native language and receive clear guidance.

Challenges I Faced & Overcame

Challenge:

Integrating multiple, distinct AI models (STT, NLP, Generative, ML) into a seamless and responsive user experience.

Solution:

I designed and implemented a microservices architecture. An AI backend (FastAPI) handled all intelligent processing, while the core application logic was managed by a robust Spring Boot backend made by teammates. This separation ensured scalability and maintainability.

Challenge:

Making the symptom checker accurate and available offline.

Solution:

I trained a lightweight k-NN machine learning model that was small enough to be bundled with the mobile application, allowing for offline predictions without needing to contact a server.

Technology Stack

Next.js React JavaScript Python FastAPI Scikit-learn LangChain Google Gemini API AssemblyAI Whisper Pinecone Docker WebSockets Jitsi

Other Contributors

KN
Kartik Nhm
SD
Shrihari Deshapande
Aviral Tripathi
SK
Sudhanva Kulkarni
NS
Navya Sastry