Sanjeevni-AI

A modern AI-healthcare solution for Rural India.

Comment

Inspiration

Our project was inspired by the healthcare disparities in India, aiming to make healthcare a right for all with the aim of solving these two crucial problems:

  • Shortage of medical professionals.
  • Lack of comprehensive disease outbreak data in rural regions to the government.

What it does

Sanjeevni is a Ai-healthcare software solution that is one-of-a-kind and is targeted to serve the medical demands of Indians, specifically in rural areas. It has these USPs:

  • Local language support: Sanjeevni supports most of the officially recognized local languages of India so that it can reach to every end of the country.
  • Context-maintainence: Sanjeevni remembers previous ailments and conversations with the users in an efficient way and gives personalized and almost realistic & better diagnoses of correlated diseases.
  • Patient-customer confidentiality: We have used advanced web3 technologies to make the platform fully anonymous, since private data/history of patients might be at risk.
  • Dashboard for monitoring of outspread of disease: Anonymous census of diseases and conditions will be presented in a dashboard format in order to prevent and take necessary steps by the government.

How we built it

We built Sanjeevni, as a responsive web application in React. The 3D model and the animations were utilized using blender and rendered into the application using three.js. We have used GPT3.5 turbo fine-tuned on PubMed dataset for the LLM, to ensure reliable and smooth consultancy. The contexts are stored anonymously in MongoDb. Inspite of having an Aadhar-login system, the anonimity is maintained by storing the Aadhar ID on IPFS and using the generated IPFS CID hash to tally with the UID that maps with the context database.

Challenges we ran into

Challenges we faced:

  • Voice Speech Recognition: We struggled with various ML models, but on the first day, none of them worked. Eventually, we successfully implemented voice recognition using the Webspeech API, which understands all regional languages.
  • Voice Translation: Translating from regional languages to English was a major hurdle since our health-assistant model only understands English. We overcame this challenge by integrating the Google Translate API.
  • Generating Responses with LLM: Initially, we received very generic answers from the health-assistant model. However, after fine-tuning the model with health research papers, we significantly improved its accuracy.
  • Text-to-Speech in Regional Languages: Implementing text-to-speech was also a tough task, as we needed to consider the accents of each language. We addressed this challenge by using the Google Cloud Text-to-Speech (TTS) npm library.
  • Blockchain Smart Contract Implementation for Securing User Info: Interacting with smart contracts without requiring users to create wallets posed a challenge. However, we successfully solved this problem by leveraging the Biconomy Gasless SDK toolkit.

Accomplishments that we're proud of

We were a team of 4 students, with no previous knowledge of 3D rendering and integrating it with a web-application. But we pulled off all of it, including animating, building and learning during the course of 36 hours, and integrated it with our previous knowledge of LLMs, Blockchain and Software Development as a whole.

What we learned

We learned 3D modelling, working with blender, time management and working as a team effectively.

What's next for Sanjeevni

Sanjeevni aims to further be accessible to people by means of calls as robo-calls. This will improve its reach and effectiveness in the country. We also aim to remove the dependency of closed sourced LLMs and switch to open-source and free ones.

Built With

Share this project:

Updates