Update README.md

README.md

@@ -25,9 +25,11 @@ It follows a **“look–think–predict”** reasoning paradigm that mimics hum
   Generates **heatmaps** highlighting potential artifacts and misalignments for more interpretable feedback.
 
 - **🗣️ Grounded Image Reasoning**  
-  Provides **step-by-step reasoning** explaining the evaluation — enhancing interpretability and trustworthiness.
+  Provides **step-by-step reasoning** that explains how the model arrives at its evaluation.  
+  ImageDoctor not only describes *why* an image succeeds or fails but also **zooms in** on flawed regions—offering spatially grounded visual evidence that highlights artifacts or misalignments.  
+  This localized reasoning enhances interpretability, transparency, and trustworthiness.
 
-- **⚙️ Reinforcement-Tuned with GRPO**  
+- **⚙️ Reinforcement-Finetuned with GRPO**  
   Trained with **Group Relative Policy Optimization (GRPO)** for richer, more stable preference alignment.
 
 - **💡 Versatile Applications**