Inspiration
The inspiration behind this project was the need to address the critical healthcare challenge of providing basic medical services to the nearly 3 billion people worldwide without internet access. We saw an opportunity to leverage AI-powered technology to create a text-based tool that would be accessible to anyone, anywhere, even those with basic flip phones. By offering free basic diagnosis through a simple text message, we aimed to reduce the burden of costly doctor visits and make healthcare more affordable and accessible to all.
What it does
This project is an innovative AI-powered text-based tool that provides free basic medical diagnosis to individuals without internet access. Patients can simply text a list of symptoms they are experiencing and receive a prompt response via SMS, allowing them to receive basic medical help from anywhere, at any time, even with basic flip phones. By reducing the need for costly doctor visits, our tool makes healthcare more accessible and affordable to all.
How we built it
Do you know how we can teach computers to understand human language (natural language)? We use something called "global vectors" to help them. It's like a special code that represents words in a way that computers can understand.
These codes are so cool because they show us different ways that words are related to each other. Like, "chest pain" and "knee pain" are related because they're both types of pain, and the code shows us that!
Now, we want to use these codes to help computers learn how to detect diseases from the SMS test. To do this, we give the computer examples of text that are either talking about a disease or not. But since we don't speak every language in the world (I wish we did!) , so use Google Translate to help us label the examples in many different languages (lenguas, langues...).
Then, we teach the computer using a special program called an "Naive Bayes classifier" to tell what disease a piece of text is probably talking about.
Challenges we ran into
We encountered some issues with Microsoft Azure and Google Cloud, as we were unable to get GPU-enabled VMs for our project due to lack of sufficient credits. As a result, training our model took an extremely long time due to the absence of CUDA functionality, something which requires an NVIDIA GPU and it would have helped speed up the compute-intensive application. Additionally, we faced issues with the VM, as it frequently started up but kept on loading. Azure also sometimes took a long time to update the firewall rules. Another issue we faced was that Microsoft's BioGPT, GPT for biomedical applications, while trained on a large corpus of biological literature was unable to produce concise advice for user queries and ChatGPT got answers often confidentally wrong.
Accomplishments that we're proud of
We are proud of several accomplishments that we have achieved with our AI-powered text-based tool. Firstly, we are happy to be helping people, especially those who do not have access to basic healthcare facilities, by providing free basic medical diagnosis via SMS. Secondly, we are proud of working effectively as a team, even with first-time hackers, to develop and deliver our solution to market successfully. Lastly, we are pleased to report that our AI's prediction accuracy is decent, even without the desired data. These accomplishments have motivated us to continue our mission of making healthcare more accessible and affordable to all.
What we learned
Through this project, we learned that AI-powered technology can be a powerful tool in addressing critical healthcare challenges. By leveraging natural language processing and text-based communication, we were able to create a simple and accessible tool that provides free basic medical diagnosis to individuals without internet access. We also learned the importance of privacy-friendly solutions and the need to ensure that our tool was accessible to everyone, including those with basic flip phones. Overall, this project taught us the value of using technology to make healthcare more accessible and affordable to all. Another thing we learnt is that documentation is sooo important. Twilio Docs (https://www.twilio.com/docs) saved us!
What's next for MedText
If we continue working on this platform, our next steps for MedTextwould be to:
- expand our tool's capabilities and scale its impact
- add more advanced diagnostic capabilities, such as identifying potential diseases and providing actionable insights to the user
- speak with medical professionals and influencers to expand the reach of our tool
- better understanding of natural language to improve our diagnosis.
Log in or sign up for Devpost to join the conversation.