About the Project: BioSentinel AI

What Inspired Us

The staggering statistic that 500,000 Americans are infected annually due to improperly handled medical waste was a wake-up call. We realized that the root cause wasn’t just the waste itself but the lack of accessible, real-time guidance for workers handling it. This inspired us to create BioSentinel AI, a tool that bridges the gap between technology and safety in medical waste management.

What We Learned

• The scale of the problem: From needlestick injuries to environmental pollution, the consequences of mismanaged waste are far-reaching.
  
• The power of AI: Custom-trained models like DETR (Detection Transformer) can accurately detect and classify medical waste in real-time.
  
• The human factor: Workers often lack proper training, and even small mistakes can have catastrophic consequences.
  

How We Built It

1. Data Collection: Downloaded and preprocessed a high-quality dataset of medical waste images (syringes, vials, masks, etc.).
   
2. Model Training: Custom-trained DETR from scratch to ensure high accuracy and adaptability for medical waste detection.
   
3. Integration: Built a user-friendly interface that overlays bounding boxes and labels on detected waste items.
   
4. Guidance System: Added a recommendation engine that provides step-by-step disposal instructions for each detected item.
   

Pipeline of the Project

1. Training the Model:
   
   • Custom-trained DETR on a medical waste dataset to detect and classify objects like syringes, vials, and masks.
     
2. Object Detection:
   
   • Used the trained model to detect objects in a random image, generating bounding boxes and labels for each item.
     
3. Custom Prompt Generation:
   
   • Extracted the detected objects and passed them into a custom prompt template.
     
4. Gemini API Integration:
   
   • Sent the custom prompt to Gemini via its API to generate detailed instructions, hazards, and recommendations.
     
5. Final Output:
   
   • Displayed the detected objects alongside Gemini’s response, including:
     
     • Disposal Instructions: Step-by-step guidance for safe handling.
       
     • Potential Hazards: Risks of improper disposal (e.g., infections, environmental damage).
       
     • Expert Recommendations: Best practices for compliance and safety.

Challenges We Faced

• Dataset Limitations: While we had a pre-downloaded dataset, ensuring it was diverse and representative of real-world scenarios required careful preprocessing.
  
• Training Complexity: Training DETR from scratch was computationally intensive and required fine-tuning to achieve optimal performance.
  
• Real-World Application: Adapting the system for use in noisy, real-world environments (e.g., cluttered trash bags) was a significant hurdle.
  

The Impact

BioSentinel AI isn’t just a tool—it’s a movement. By providing real-time guidance, we’re empowering workers, preventing infections, and protecting the environment. Every scan is a step toward a safer, healthier future.

BioSentinel AI: Because every piece of waste tells a story—let’s make sure it’s a safe one. 💉🌍

Built With

• fastapi
• gemini
• python
• pytorch
• rest
• streamlit