What AI says about LLMCoder

LLMCoder is a framework that uses domain-specific fine-tuned large language models to automate ICD-10 medical coding. It employs a two-stage fine-tuning pipeline—initial training on the full ICD-10 code set followed by targeted fine-tuning for linguistic variations—achieving high accuracy on real-world clinical notes. It supports both proprietary (GPT-4o mini) and open-source (Llama) models and is optimized for local deployment in resource-constrained environments (e.g., requiring only 2GB GPU memory).

Hosted on GitHub