added a streaming Audio input component per the Gradio guide—using Whisper for transcription in real time and piping the resulting text into the chatflow

src/app.py

@@ -1,11 +1,9 @@
 import os
 import gradio as gr
+import openai
 from transformers import pipeline
 from llama_index import SimpleDirectoryReader, GPTVectorStoreIndex
-from llama_index.llm_predictor import HuggingFaceLLMPredictor, LLMPredictor
-
-# Optional OpenAI import remains for default predictor
-import openai
+from llama_index.llm_predictor import HuggingFaceLLMPredictor
 
 # --- Whisper ASR setup ---
 asr = pipeline(
@@ -26,49 +24,53 @@ or, if you have enough info, output a final JSON with fields:
 {"diagnoses":[…], "confidences":[…]}.
 """
 
+def transcribe_and_respond(audio_chunk, state):
+    # Transcribe audio chunk
+    result = asr(audio_chunk)
+    text = result.get('text', '').strip()
+    if not text:
+        return state, []
 
-def transcribe_and_respond(audio, history):
-    # 1) Transcribe audio → text
-    user_text = asr(audio)["text"]
-    history = history or []
-    history.append(("user", user_text))
-
-    # 2) Build unified prompt for LLM
-    messages = [("system", SYSTEM_PROMPT)] + history
-    prompt = "\n".join(f"{role.capitalize()}: {text}" for role, text in messages)
-    prompt += "\nAssistant:"
+    # Append user message
+    state.append(("user", text))
 
-    # 3) Select predictor (OpenAI or Mistral/local)
-    predictor = get_llm_predictor()
-    resp = predictor.predict(prompt)
+    # Build LLM predictor (you can swap OpenAI / HuggingFace here)
+    llm_predictor = HuggingFaceLLMPredictor(model_name_or_path=os.getenv("HF_MODEL", "gpt2-medium"))
 
-    # 4) If JSON-style output, treat as final
-    if resp.strip().startswith("{"):
-        result = query_symptoms(resp)
-        history.append(("assistant", f"Here is your diagnosis: {result}"))
-        return "", history
+    # Query index with conversation
+    # (Assuming `symptom_index` is your GPTVectorStoreIndex)
+    # Prepare combined prompt from state
+    prompt = "\n".join([f"{role}: {msg}" for role, msg in state])
+    response = symptom_index.as_query_engine(
+        llm_predictor=llm_predictor
+    ).query(prompt)
+    reply = response.response
 
-    # 5) Otherwise, it's a follow-up question
-    history.append(("assistant", resp))
-    return "", history
+    # Append assistant message
+    state.append(("assistant", reply))
 
+    # Return updated state to chatbot
+    return state, state
 
-# --- Build Gradio app ---
-with gr.Blocks() as demo:
-    gr.Markdown("## Symptom to ICD-10 Diagnoser (audio & chat)")
+# Build Gradio interface
+demo = gr.Blocks()
+with demo:
+    gr.Markdown("# Symptom to ICD-10 Code Lookup (Audio Input)")
     chatbot = gr.Chatbot(label="Conversation")
-    mic     = gr.Microphone(label="Describe your symptoms")
-    state   = gr.State([])
+    state = gr.State([])
+    # Use streaming audio input for real-time transcription
+    mic = gr.Audio(source="microphone", type="filepath", streaming=True, label="Describe your symptoms")
 
-    mic.submit(
+    mic.stream(
         fn=transcribe_and_respond,
         inputs=[mic, state],
-        outputs=[mic, chatbot, state]
+        outputs=[chatbot, state],
+        time_limit=60,
+        stream_every=5,
+        concurrency_limit=1
     )
 
 if __name__ == "__main__":
     demo.launch(
-        server_name="0.0.0.0",
-        server_port=7860,
-        mcp_server=True
+        server_name="0.0.0.0", server_port=7860, mcp_server=True
     )