Update

README.md

@@ -13,79 +13,54 @@ secrets:
 
 # 🧬 MediAgent AI: Your Conversational Health Navigator
 
-**MediAgent AI** is a sophisticated, conversational health assistant designed to provide reliable, sourced, and easy-to-understand answers to a wide range of medical questions. It uses a state-of-the-art, two-stage process to first understand a user's needs through conversation and then use powerful tools to find an evidence-based answer.
+**MediAgent AI** is a sophisticated, conversational health assistant designed to provide reliable, sourced, and easy-to-understand answers to a wide range of medical questions.
 
 This application is built with a decoupled architecture: a modern Gradio UI hosted on Hugging Face Spaces and a powerful, serverless GPU backend running on [Modal](https://modal.com).
 
 ## 🚀 Key Features
 
-- 💬 **Conversational & Stateful**: Instead of a simple Q&A, MediAgent engages in a multi-turn dialogue to gather necessary details, ensuring the final answer is relevant and comprehensive.
-- 👤 **Role-Based Adaptation**: The AI tailors its language and questions based on the selected user role: **Patient** (simple, empathetic language), **Doctor** (technical, concise), or **Health Worker** (procedural, clear).
-- 🛠️ **Agentic Tool Use**: After gathering enough information, the agent actively chooses the best tool for your question from its specialized toolkit:
-    - `search_pubmed`: For technical and clinical research.
-    - `search_web`: For general health and symptom information.
-    - `check_drug_interactions`: To check for interactions between multiple drugs via the openFDA API.
-    - `check_drug_side_effects`: To find adverse reactions for a specific drug via the openFDA API.
-- 🛡️ **Advanced Safety Layers**: Features critical safety prompts that instruct the AI on how to handle potential emergencies and always provide disclaimers.
-- 🧠 **Two-Stage Reasoning Engine**: Powered by **`google/gemma-2-2b-it`** and LlamaIndex, the system first uses a `StructuredOutputManager` to gather information conversationally, then deploys a `ReActAgent` to think, act, and observe in a loop to find the final answer.
-- ✨ **Modern Chat UI**: A clean, responsive chat interface built with [Gradio](https://www.gradio.app) that supports natural, multi-turn conversations.
+- 🌐 **Multilingual Support**: Automatically detects and responds in English, French, Spanish, and Hindi.
+- 💬 **Conversational & Stateful**: Remembers the chat history to understand follow-up questions.
+- 👤 **Role-Based Adaptation**: Tailors its language and the tools it uses based on whether you select **Patient** or **Doctor**.
+- 🛠️ **Multi-Tool Capability**: Uses a powerful toolkit to search the web, PubMed, clinical trials, and drug databases to find the best answer.
+- 🛡️ **Safety First**: Includes a critical safety layer to detect potential emergencies and provide appropriate disclaimers.
 
 ---
 
-## 🧪 Example Conversation (Patient Role)
-
-> **You:** I have a headache.
->
-> **MediAgent AI:** I'm sorry to hear that. How long have you had this headache?
->
-> **You:** For about two days, and it's pretty severe.
->
-> **MediAgent AI:** To confirm, you are experiencing a **severe** headache that has lasted for **about two days**. Is that correct?
->
-> **You:** Yes.
->
-> **MediAgent AI:** (Now using its tools) Based on information from reputable medical sources, a severe headache lasting two days can be associated with several conditions... It's important to monitor for other symptoms such as... [Provides detailed, sourced information].
->
-> ---
-> _Disclaimer: This is an AI-generated answer and not a substitute for professional medical advice..._
+## 🧪 Try It In Your Language!
+
+Copy and paste one of the examples below to test the AI's multilingual capabilities.
+
+### English
+
+What are the most common side effects of metformin?
+
+### Spanish
+
+Cuáles son los síntomas de la diabetes tipo 2?
+
+### French
+
+Comment savoir si j'ai la grippe ou un simple rhume ?
+
+### Hindi
+
+उच्च रक्तचाप के लिए घरेलू उपचार क्या हैं?
 
 ---
 
-## 🛠️ Architecture Overview
-
-MediAgent AI uses an advanced two-stage architecture that separates conversational management from tool-based reasoning.
-
-```plaintext
-                   User Input + Role + State
-                               │
-              ┌────────────────┴────────────────┐
-              │  Gradio Chat UI on HF Spaces    │
-              │ (Maintains Conversation State)  │
-              └────────────────┬────────────────┘
-                               │ (API Call to Modal with full context)
-              ┌────────────────↓────────────────┐
-              │  Modal Serverless Backend       │
-              └────────────────��────────────────┘
-                               │
-           1. 🛡️ Safety & Pre-processing
-                               │
-┌─────────► 2. 🗣️ StructuredOutputManager (Conversational Loop)
-│              a. Is task ready? (e.g., has user confirmed summary?)
-│              b. If NO: Use LLM to generate a clarifying question or summary.
-│              c. Send response back to UI and wait for next user input.
-│                               |
-└───────────────────────────────┘
-                               │ (If YES, proceed to next stage)
-                               │
-           3. 🤖 LlamaIndex ReAct Agent (Tool-Using Loop)
-                  a. 🤔 Thought: "User confirmed headache for 2 days. I should use search_web."
-                  b. 🎬 Action: Executes tool (`search_web('severe headache 2 days')`)
-                  c. 👀 Observation: Receives data back from the tool (e.g., search results)
-                  (Loop continues until the agent has a complete answer)
-                               │
-           4. 📝 Final Answer Synthesis
-         (Agent composes a complete answer based on tool output)
-                               │
-              ┌─────────────────────────────┐
-              │ Final Answer Displayed in UI  │
-              └─────────────────────────────┘
+## 🧑‍⚕️ See The Role-Based Difference
+
+Select a role in the UI and use a matching prompt below to see how the AI adapts its response.
+
+### For the Patient Role
+
+Select **patient** and ask a general health question. The AI will use web search to provide an easy-to-understand overview.
+
+What are the most common side effects I should watch out for with lisinopril?
+
+### For the Doctor Role
+
+Select **doctor** and ask a technical, multi-part question. The AI will use specialized tools like PubMed and ClinicalTrials.gov to provide a detailed, sourced answer.
+
+My patient with Crohn's is on methotrexate and we are considering adding infliximab. I need the latest PubMed literature on efficacy, a check for interactions between the two drugs, and a list of ongoing clinical trials for novel biologics in IBD.

app.py

@@ -99,13 +99,27 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="sky"), cs
     </div>
     """)
 
+    with gr.Row():
+        gr.Markdown("""
+        <b>💡 Example Prompts:</b>
+        <ul>
+            <li><button onclick=\"document.querySelector('textarea').value='What are the most common side effects of metformin?';\">English: What are the most common side effects of metformin?</button></li>
+            <li><button onclick=\"document.querySelector('textarea').value='Cuáles son los síntomas de la diabetes tipo 2?';\">Spanish: Cuáles son los síntomas de la diabetes tipo 2?</button></li>
+            <li><button onclick=\"document.querySelector('textarea').value='Comment savoir si j\'ai la grippe ou un simple rhume ?';\">French: Comment savoir si j'ai la grippe ou un simple rhume ?</button></li>
+            <li><button onclick=\"document.querySelector('textarea').value='उच्च रक्तचाप के लिए घरेलू उपचार क्या हैं?';\">Hindi: उच्च रक्तचाप के लिए घरेलू उपचार क्या हैं?</button></li>
+        </ul>
+        """)
+
     with gr.Group(elem_id="role-selector"):
         user_role = gr.Radio(
-            ["patient", "doctor", "health_worker"],
+            ["patient", "doctor"],
             label="👤 Select Your Role",
             value="patient",
-            info="The AI will adapt its communication style based on your role."
+            info="The AI will adapt its tools and communication style based on your role.",
+            elem_id="user-role-radio",
+            show_label=True
         )
+        gr.Markdown("<span title='Choose your role to tailor the AI response. Patients get simple answers, doctors get technical details.' style='cursor: help;'>ℹ️</span>")
 
     chatbot = gr.Chatbot(
         label="Conversation",
@@ -123,6 +137,7 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="sky"), cs
             container=False,
         )
         submit_btn = gr.Button("Send", variant="primary", scale=1, min_width=150, elem_id="send-button")
+        loading_spinner = gr.Markdown("<span id='loading-spinner' style='display:none;'>⏳ <em>Thinking...</em></span>")
 
     with gr.Row():
         clear_btn = gr.ClearButton(
@@ -131,24 +146,26 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="sky"), cs
         )
 
     def handle_user_turn(user_input, chat_history, user_role_selection, current_state, sid):
+        import time
+        from gradio import processing
+        # Show spinner
+        gr.update(elem_id="loading-spinner", value="<span id='loading-spinner' style='display:block;'>⏳ <em>Thinking...</em></span>")
         if not user_input.strip():
+            gr.update(elem_id="loading-spinner", value="<span id='loading-spinner' style='display:none;'></span>")
             yield {chatbot: chat_history, conversation_state: current_state, user_textbox: user_input}
             return
-
         user_message_html = f"<div class='message user'>{user_input}</div>"
         chat_history.append((user_message_html, None))
         yield {chatbot: chat_history, user_textbox: ""}
-
         if not MODAL_BACKEND_URL:
             error_message = "### ❗ Configuration Error\n**Reason**: `MODAL_BACKEND_URL` environment variable is not set."
             error_message_html = f"<div class='message bot'>{error_message}</div>"
             chat_history[-1] = (user_message_html, error_message_html)
+            gr.update(elem_id="loading-spinner", value="<span id='loading-spinner' style='display:none;'></span>")
             yield {chatbot: chat_history}
             return
-
         headers = {"Content-Type": "application/json"}
         payload = {"user_input": user_input, "session_id": sid, "role": user_role_selection}
-
         try:
             response = requests.post(MODAL_BACKEND_URL, headers=headers, json=payload, timeout=300)
             response.raise_for_status()
@@ -156,9 +173,9 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue", secondary_hue="sky"), cs
             bot_response_text = backend_response.get("final_markdown", "Sorry, I received an empty response.")
         except requests.exceptions.RequestException as e:
             bot_response_text = f"### ❗ Network Error\n**Reason**: Could not connect to the backend.\n**Details**: {str(e)}"
-
         bot_message_html = f"<div class='message bot'>{bot_response_text}</div>"
         chat_history[-1] = (user_message_html, bot_message_html)
+        gr.update(elem_id="loading-spinner", value="<span id='loading-spinner' style='display:none;'></span>")
         yield {chatbot: chat_history, conversation_state: current_state}
 
     submit_btn.click(