TransfuSense

• 

• 

• 

• 

• 

• 

Inspiration Every 3–4 weeks, millions of Thalassemia patients rely on blood transfusions to survive. But their journey is filled with uncertainty — interpreting complex lab reports, predicting next transfusion dates, and accessing multilingual medical guidance is a daunting task, especially in underserved regions.

We asked ourselves: 🔴 In a world powered by AI, why is life-saving care still running on paper and panic?

This led us to build TransfuSense — an AI-powered assistant to make critical care smarter, faster, and more accessible.

💡 What it does TransfuSense is a multilingual AI assistant designed to streamline the Thalassemia care cycle. It:

📄 Reads and understands lab reports using OCR and NLP

🧠 Summarizes key medical details in natural language

📅 Predicts next blood transfusion dates using AI models

🌐 Translates insights into the user's preferred language

💬 Acts as an interactive FAQ bot for common patient queries

All accessible through a clean Streamlit interface with modular tools.

🏗️ How we built it 👁 OCR using Tesseract to extract text from scanned lab reports

🧾 NLP-powered summarizer to parse medical information

📈 Predictive model trained to estimate transfusion timelines based on hemoglobin and historical data

🤖 Intent classification for dynamic routing of user queries

🌍 Multilingual translation using OpenAI models

🧠 Streamlit-based UI with tabbed tools for accessibility

🗃 Memory management and session tracking via a custom controller-agent pipeline

🧱 Challenges we ran into 🔍 Extracting structured data (e.g., hemoglobin level, date) from unstructured OCR output

🗣️ Ensuring accurate translations across medical terminology

🤹 Integrating modular tools under a unified intelligent assistant

📐 Calibrating prediction models with limited real-world datasets

🧼 UI/UX design to make the platform intuitive yet powerful

🏅 Accomplishments that we're proud of 🔧 Built an end-to-end AI assistant in under 48 hours

🧠 Created a working demo that extracts → summarizes → predicts from real lab reports

🌐 Made it accessible across languages for diverse users

📸 Seamlessly linked OCR, NLP, and ML tools in a single workflow

📚 What we learned How to build multi-stage pipelines with OCR + NLP + prediction logic

Importance of context preservation in multilingual healthcare

How modularity helps scale and debug AI systems under time constraints

Real-life complexities of medical document analysis

🚀 What's next for TransfuSense 🧬 Integrate with hospital databases for real-time lab report access

🔗 Connect with blood bank APIs for donor availability tracking

📲 Launch as a WhatsApp-based bot for rural accessibility

🧑‍⚕️ Add clinician-mode for doctors to monitor patient timelines

🔍 Include explainable AI modules for prediction transparency

Built With

• chromadb
• distilbart
• easyocr
• flask
• langchain
• mistral
• nllb
• numpy
• pandas
• python
• scikit-learn
• streamlit