Sanjeevni-AI

GitHub Repository
Owner: Deepanshu Tyagi

Inspiration

Sanjeevni was inspired by the healthcare disparities in India, with a mission to make healthcare a right for all. This project aims to address two critical issues:

• Shortage of medical professionals.
• Lack of comprehensive disease outbreak data in rural areas for the government.

What It Does

Sanjeevni is an AI-powered healthcare software solution designed specifically for the medical needs of Indians, especially in rural regions. Key features include:

• Local Language Support: Supports most of India's officially recognized local languages to reach all corners of the country.
• Context Maintenance: Remembers previous ailments and user interactions to provide personalized, accurate diagnoses of related health issues.
• Patient-Confidentiality: Uses advanced Web3 technologies to ensure anonymity and protect private patient data.
• Disease Outbreak Monitoring Dashboard: Provides a dashboard with anonymized data on disease trends to aid government response and preventive actions.

How We Built It

Sanjeevni is a responsive web application built with:

• Frontend: React, with 3D models and animations created in Blender and integrated using Three.js.
• LLM: GPT-3.5-turbo fine-tuned on PubMed data, ensuring reliable and effective consultancy.
• Database: MongoDB for storing contexts anonymously.
• Aadhar Login Anonymity: Aadhar IDs are stored on IPFS, with IPFS CIDs used to manage user sessions without direct personal identifiers.

Challenges We Faced

1. Voice Speech Recognition: Initially faced issues with ML models, later resolved using the WebSpeech API for regional language support.
2. Voice Translation: Overcame language translation challenges by integrating the Google Translate API to translate regional languages to English.
3. LLM Response Generation: Improved health assistant model accuracy by fine-tuning with health research papers.
4. Text-to-Speech in Regional Languages: Addressed language accent challenges by using the Google Cloud Text-to-Speech (TTS) library.
5. Blockchain Smart Contract for User Security: Managed gasless smart contract interaction using Biconomy SDK, avoiding the need for users to create wallets.

Accomplishments

With a team of two students with no prior experience in 3D rendering, we achieved full integration of 3D animations, LLMs, blockchain, and web application development within 36 hours.

Lessons Learned

• 3D modeling and Blender
• Effective team collaboration
• Time management under a tight deadline

Future Plans

• Voice Accessibility: Enable access via automated calls, improving reach.
• Open-Source LLMs: Transition from closed to open-source LLMs to reduce dependency on proprietary models.

Built With

• Blender
• Ethereum
• Express.js
• IPFS
• MongoDB
• OpenAI
• React
• Three.js