-
-
The main page of the app
-
Live transcription gets recorded here. I have inputted a pre-recorded transcript just for Demo.
-
The final output after processing is done.
Inspiration
What it does
How we built it
Challenges we ran into
Inspiration
My mother is a doctor and she has her own clinic. She's a really really good doctor and hence she has to see a huge number of patients every day. Although she has so many patients to see, she ensures that she spends a good amount of time with them instead of rushing through them. Due to this, a lot of time goes in writing, formatting and printing a prescription with all the past history of the patient etc. Also, my mom's handwriting is not the best, so it becomes difficult for patients to understanding the handwriting on her prescriptions post consult (this happens with many doctors actually)
What it does
The app basically listens in on a doctor-patient consultation. The doctor presses "Record COnversation" only after the patient consents to recording. Then throughout the consult, a live transcript is generated of the conversation. At the end, the doctor has to press "Stop and Generate prescription". Then, an AI Model (i am using Gemini 2.5-Pro) will process the transcript and then return the final prescription which contains all the necessary details of the consult and excludes the irrelevant details.
How we built it
I started off by first designing the User interface - wanted a general idea of what it should look like. Then started working on the front end designed in Basic HTML and CSS. Backend was developed in JavaScript, and it took a long time to perfect the prompt. Now, it's ready!
Challenges we ran into
The one thing is that transcription service is NOT able toaccurately recognise the names of various Indian drugs. For example, there is a medicine called "Econorm" used for treating loose motions and is widely used in India. However, when the word "Econorm" is used in the conversation, the transcript detects it as "Economic" or sometimes "Economical" which becomes a problem. Otherwise, it has been found to be INCREDIBLY accurate.
Accomplishments that we're proud of
This app has been incredibly helpful to my mother - she's the only doctor I have tried this with. The accuracy has truly been unbelievable in the pilot test as well. There are refinements which can be done and it should be done if this has to be production ready. Also, the transcriber can accurately detect text which is spoken in a different accent. I am studying in the US now, and people who have an accent when they speak is traditionally a challenge to transcribe but it still worked perfectly!
## What we learned Well, we need to train the model to recognise various Indian drugs if it truly has to be production ready. Also, I have learned that atleast in India, this app has HUGE potential to be successful and be present in every single clinic across the country.
What's next for preSCRIBE - Linking Conversation And Care
Again, work has to be done to ensure accuracy in the service and it must be updated regularly. There is lots of scope for commercialization in India and probably in other countries as well.
Log in or sign up for Devpost to join the conversation.