DocMate

• 

  A chat Interface

• 

  Proof for utilization of Intel Xeon With AMX

Inspiration

Inspiration for this project is that people often have few minor symptoms they are facing, while they doesn't want to visit doctor directly they may go to search engines to get any info. The most of the top suggestions suggested by these search engines are very much categorized to dangerous disease which may worry user. My application was specifically designed to address this issue.

What it does

The application considers the users symptoms and predicts most likely disease from it. It also provides additional symptoms that user can counter verify if they have these additional symptoms or not. Finally our application provides remedies for temporary relief for the user's symptoms. The other thing it is capable of is it can handle queries that requires latest medical treatments.

How we built it

I have considered Gemma 2B- It model from hugging face for this which was designed by google. I have compressed the model using OpenVino and converted the model into OpenVino IR. I have considered 2 Precision models for processing query based on the intent of the user's prompt. For the RAG approach I have utilized PubMed search which can help in providing latest and up to date information in medical. I have considered intent based classification for user's input which helps to make efficient handling of user's request. Finally for chatbot I have used Gradio which can launch the application through Jupyter notebook and deployed through OpehShiftAI environment

Challenges we ran into

The main challenge is prompt engineering since using a relatively small model for this task would actually make the model to generate inconsistent results at times.

Accomplishments that we're proud of

The accomplishments I can proud is that it could provide accurate responses most time in a concise manner. This helps the application to serve as Level 1 support for user's in medical aspects. The RAG approach maintains the up to date retrieval of information which the user can consider if he/she has any chronic condition

What we learned

I personally learned a lot in terms of model compressions by OpenVino, getting used to OpenShift AI environment, RAG approach through PubMed. This learning helps me to come up with more exciting use cases in future

What's next for DocMate

The next for DocMate is to handle medical images, handle user sessions in terms of remembering the user and respond relevant to the previous encounter. This can be accurately done by using a bit more large models and processing power. Would be more exciting if I was able to add Vision and Speech with emotion which will make it a complete application.

Built With

• gemma-2b
• gradio
• openshiftai
• openvino