Hemo

Hemo is your personal Blood Health & Blood Donation Assistant.

Comment

Inspiration

Last year I had to run to find blood for my sick family members. When faced with this issue, I saw that most developing countries including mine don't have proper blood donation network facilities. There may be hundreds of voluntary blood donor groups, but only a small percentage of them actually donate regularly. Hence I made Hemo, an AI Blood Health & Donation assistant.

What it does

Hemo has the following features

  • Hemo helps you learn more about your blood report through open source AI language model. You can upload & chat with your report.
  • If you run a blood donation center & need to analyze the blood donation pattern of your donors, Hemo can help you! You can upload a CSV file of the Blood Donation info & Hemo will automatically create a pairplot to visualize the blood donation patterns.
  • Find nearby blood donation centers by just entering your current location & see them on the map.

How we built it

Hemo is built using streamlit & it is fully open source, as it does not use any OpenAI API & instead uses a free & open source language model Model Specs: Model Name: impira/layoutlm-doc-vqa Model Type: document-question-answering

I imported the model using hugging face transformers. Then I use streamlit to design the UI elements like input form & sidebar file upload.Firstly you can upload an image of your blood-test report. It has to be a jpg,png format. Then you can ask the Language Model questions about your report.

On the sidebar you can uplaod CSV data of patients blood donation history to create a plot of blood donation patterns. I used Pandas Library for prediction of blood patterns & uses streamlit to display the plot.

Finally donors can search for nearby donation centers. On the sidebar, their is option to enter your current location. I used Google Places API for this feature. You can enter your location (ex: New York) & it will show all available healthcare centers where you can donate blood within a 5 KM radius.

Challenges we ran into

This was my first time building a feature rich AI app using streamlit. Also I had '0' idea about using open source models. But I was kinda sick of having to create a new API key from OpenAI everytime I tried to build an AI app. So I forced myself to learn how Hugging Face AI open source models work. I used a smaller model. But it does the job just fine.I didn't have the Tesseract-OCR installed on my device which was necessary to convert the uploaded images into text as the model I used only excepts image files. So figuring that out & configuring Tesseract-OCR was a big challenge.

I have never used Google Places API before or any Google API for that matter. So figuring out how to use the API was a challenge. Specially on how to convert the text values into latitude & longtitude so the API could send back the right json response.

Accomplishments that we're proud of

I learned so much through this project. I feel very accomplished to have built my first ever AI web app using streamlit. I wouldn’t say the journey was easy.But if you keep at it, you can basically figure out any coding problem, that's what I learned.

What we learned

I learned about using streamlit's many awesome features. Specially the st.map & st.plot features which helped me visualize the data. Also learned many features of the Pandas Library which is the backbone of the blood donation pattern prediction.But most of all I learned a lot about open source LLMs, this project introduced me to the huge world of open-source LLMs.

What's next for Hemo

Ultimately Hemo aims to encourage the increase pf blood donation. Hemo aims to encourage people to donate their blood by informing & educating them on blood donation & transfusion methods. I also aim to use a more powerful open source LLM in the future like llama or Mistral so that Hemo can provide more expert level advice on your Blood Health.

Built With

Share this project:

Updates