MEDI MIND AI is a transformative AI tool developed to help healthcare professionals tackle the costly issue of misclassified Diagnosis-Related Group (DRG) codes, while enhancing workflow through live transcription, medical entity extraction, and patient history summarization.
Using cutting-edge models—BERT, LLAMA2, and QWEN2—our approach included:

Data Collection and Processing: Acquiring secure, compliant data for processing.

Model Training and Optimization: Achieving over 86% accuracy in top 10 DRG code predictions, setting a new benchmark.

Feature Development: Incorporating real-time transcription, entity extraction, and summarization.

Testing and Deployment: Conducting thorough testing to ensure clinical reliability.

One challenge was the model’s limited input size, which affected accuracy due to the absence of CUDA drivers essential for fine-tuning. Despite this, the model has impressed our client, promising even greater accuracy with next-gen tools, paving its way to real-world application.

MEDI MIND AI represents a new era in medical coding, poised to save hospitals millions and free up essential hours. Beyond the numbers, this project revealed the extraordinary versatility of Large Language Models (LLMs) and their potential to reshape healthcare, reducing administrative stress so healthcare providers can focus on patient care.