no need for gradio live

app.py

@@ -2,4 +2,8 @@
 from src.app import demo
 
 if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", server_port=7860)
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+        show_api=True  # Shows the API documentation
+    )

src/app.py

@@ -66,43 +66,22 @@ def process_audio(audio_array, sample_rate):
     if audio_array.ndim > 1:
         audio_array = audio_array.mean(axis=1)
     
-    # Normalize audio
-    audio_array = audio_array.astype(np.float32)
-    audio_array /= np.max(np.abs(audio_array))
+    # Convert to tensor for resampling
+    audio_tensor = torch.FloatTensor(audio_array)
     
     # Resample to 16kHz if needed
     if sample_rate != 16000:
-        resampler = T.Resample(orig_freq=sample_rate, new_freq=16000)
-        audio_tensor = torch.FloatTensor(audio_array)
+        resampler = T.Resample(sample_rate, 16000)
         audio_tensor = resampler(audio_tensor)
-        audio_array = audio_tensor.numpy()
     
-    # Process with correct input format
-    inputs = processor(
-        audio_array, 
-        sampling_rate=16000, 
-        return_tensors="pt"
-    )
+    # Normalize
+    audio_tensor = audio_tensor / torch.max(torch.abs(audio_tensor))
     
+    # Convert back to numpy array and return in correct format
     return {
-        "input_features": inputs.input_features,
-        "attention_mask": inputs.attention_mask
-    }
-
-# Update transcriber configuration
-transcriber = pipeline(
-    "automatic-speech-recognition",
-    model="openai/whisper-base.en",
-    chunk_length_s=30,
-    stride_length_s=5,
-    device="cpu",
-    torch_dtype=torch.float32,
-    feature_extractor=feature_extractor,
-    generate_kwargs={
-        "use_cache": True,
-        "return_timestamps": True
+        "raw": audio_tensor.numpy(),  # Key must be "raw"
+        "sampling_rate": 16000        # Key must be "sampling_rate"
     }
-)
 
 def get_system_specs() -> Dict[str, float]:
     """Get system specifications."""
@@ -312,14 +291,6 @@ def process_speech(audio_data, history):
         print(f"Processing error: {str(e)}")
         return []
 
-def update_transcription(audio_path):
-    """Update transcription box with speech recognition results."""
-    if not audio_path:
-        return ""
-    # Extract transcription from audio result
-    transcript = audio_path[1] if isinstance(audio_path, tuple) else audio_path
-    return transcript
-
 # Build enhanced Gradio interface
 with gr.Blocks(
     theme="default",
@@ -332,7 +303,9 @@ with gr.Blocks(
             font-family: ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, 
                         'Liberation Mono', 'Courier New', monospace;
         }
-    """
+    """,
+    analytics_enabled=True,
+    title="MedCode MCP",
 ) as demo:
     gr.Markdown("""
     # 🏥 Medical Symptom to ICD-10 Code Assistant
@@ -509,16 +482,10 @@ with gr.Blocks(
         # Normalize
         audio_tensor = audio_tensor / torch.max(torch.abs(audio_tensor))
         
-        # Use feature extractor with correct sampling rate
-        features = feature_extractor(
-            audio_tensor.numpy(),
-            sampling_rate=16000,  # Always use 16kHz
-            return_tensors="pt"
-        )
-        
+        # Convert back to numpy array and return in correct format
         return {
-            "input_features": features.input_features,
-            "sampling_rate": 16000  # Return resampled rate
+            "raw": audio_tensor.numpy(),  # Key must be "raw"
+            "sampling_rate": 16000        # Key must be "sampling_rate"
         }
 
     # Update transcription handler
@@ -527,17 +494,16 @@ with gr.Blocks(
         if not audio or not isinstance(audio, tuple):
             return ""
         
+        try:
             sample_rate, audio_array = audio
             features = process_audio(audio_array, sample_rate)
             
-            # Get pipeline and transcribe
             asr = get_asr_pipeline()
             result = asr(features)
             
-            if isinstance(result, dict):
-                return result.get("text", "").strip()
-            elif isinstance(result, str):
-                return result.strip()
+            return result.get("text", "").strip() if isinstance(result, dict) else str(result).strip()
+        except Exception as e:
+            print(f"Transcription error: {str(e)}")
             return ""
 
     microphone.stream(
@@ -566,30 +532,30 @@ with gr.Blocks(
         if not text:
             return history
 
-            # Limit input length
-            if len(text) > 500:
-                text = text[:500] + "..."
-                
-            # Process the symptoms
-            diagnosis_query = f"""
-            Based on these symptoms: '{text}'
-            Provide relevant ICD-10 codes and diagnostic questions.
-            Focus on clinical implications.
-            Limit response to 1000 characters.
-            """
-            response = symptom_index.as_query_engine().query(diagnosis_query)
+        # Limit input length
+        if len(text) > 500:
+            text = text[:500] + "..."
             
-            # Clean up memory
-            cleanup_memory()
-
-            return history + [
-                {"role": "user", "content": text},
-                {"role": "assistant", "content": format_response_for_user({
-                    "diagnoses": [],
-                    "confidences": [],
-                    "follow_up": str(response)[:1000]  # Limit response length
-                })}
-            ]
+        # Process the symptoms
+        diagnosis_query = f"""
+        Based on these symptoms: '{text}'
+        Provide relevant ICD-10 codes and diagnostic questions.
+        Focus on clinical implications.
+        Limit response to 1000 characters.
+        """
+        response = symptom_index.as_query_engine().query(diagnosis_query)
+        
+        # Clean up memory
+        cleanup_memory()
+
+        return history + [
+            {"role": "user", "content": text},
+            {"role": "assistant", "content": format_response_for_user({
+                "diagnoses": [],
+                "confidences": [],
+                "follow_up": str(response)[:1000]  # Limit response length
+            })}
+        ]
 
     submit_btn.click(
         fn=process_text_input,
@@ -617,10 +583,13 @@ with gr.Blocks(
     - Sharing this tool with others in healthcare tech
     """)
 
-if __name__ == "__main__":
-    demo.launch(
-        server_name="0.0.0.0",
-        server_port=7860,
-        mcp_server=True,
-        allowed_paths=["*"]
-    )
+def process_symptoms(symptoms: str):
+    """Convert symptoms to ICD codes using the configured LLM"""
+    try:
+        # Use the configured LLM to process symptoms
+        response = llm.complete(
+            f"Convert these symptoms to ICD-10 codes: {symptoms}"
+        )
+        return {"icd_codes": response.text, "status": "success"}
+    except Exception as e:
+        return {"error": str(e), "status": "error"}