Update app.py

app.py

@@ -1,7 +1,3 @@
-# Note: For Huggingface Spaces, ensure the Dockerfile includes:
-# RUN mkdir -p /tmp/cache/
-# RUN chmod a+rwx -R /tmp/cache/
-# ENV TRANSFORMERS_CACHE=/tmp/cache/
 import os
 import json
 import re
@@ -38,11 +34,15 @@ logging.basicConfig(
 )
 logger = logging.getLogger(__name__)
 
-# Set up environment for Huggingface Spaces
-OUTPUT_DIR = "/data/outputs"  # Persistent storage in Huggingface Spaces
+# Set up environment
+# For Huggingface Spaces, use /tmp for temporary storage
+if os.path.exists("/tmp"):
+    OUTPUT_DIR = "/tmp/outputs"  # Use /tmp for Huggingface Spaces
+else:
+    OUTPUT_DIR = os.path.join(os.getcwd(), "outputs")  # Fallback for local dev
 os.makedirs(OUTPUT_DIR, exist_ok=True)
+logger.info(f"Using output directory: {OUTPUT_DIR}")
 os.environ["COQUI_TOS_AGREED"] = "1"
-gr.set_static_paths(paths=[OUTPUT_DIR])  # Expose OUTPUT_DIR for file access
 
 # Define Pydantic model for slide data
 class Slide(BaseModel):
@@ -145,9 +145,14 @@ def clean_script_text(script):
 
 # Helper function to validate and convert speaker audio (MP3 or WAV)
 async def validate_and_convert_speaker_audio(speaker_audio):
-    if not os.path.exists(speaker_audio):
-        logger.error("Speaker audio file does not exist: %s", speaker_audio)
-        return None
+    if not speaker_audio or not os.path.exists(speaker_audio):
+        logger.warning("Speaker audio file does not exist: %s. Using default voice.", speaker_audio)
+        default_voice = os.path.join(os.path.dirname(__file__), "feynman.mp3")
+        if os.path.exists(default_voice):
+            speaker_audio = default_voice
+        else:
+            logger.error("Default voice not found. Cannot proceed with TTS.")
+            return None
     
     try:
         # Check file extension
@@ -157,7 +162,7 @@ async def validate_and_convert_speaker_audio(speaker_audio):
             audio = AudioSegment.from_mp3(speaker_audio)
             # Convert to mono, 22050 Hz
             audio = audio.set_channels(1).set_frame_rate(22050)
-            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_file:
+            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False, dir=OUTPUT_DIR) as temp_file:
                 audio.export(temp_file.name, format="wav")
                 speaker_wav = temp_file.name
         elif ext == ".wav":
@@ -177,7 +182,7 @@ async def validate_and_convert_speaker_audio(speaker_audio):
         if data.ndim == 2:
             logger.info("Converting stereo WAV to mono: %s", speaker_wav)
             data = data.mean(axis=1)
-            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_file:
+            with tempfile.NamedTemporaryFile(suffix=".wav", delete=False, dir=OUTPUT_DIR) as temp_file:
                 sf.write(temp_file.name, data, samplerate)
                 speaker_wav = temp_file.name
         
@@ -209,35 +214,52 @@ def extract_json_from_message(message):
         if not isinstance(content, str):
             logger.warning("TextMessage content is not a string: %s", content)
             return None
-        # Try standard JSON block
+        
+        # Try standard JSON block with triple backticks
         pattern = r"```json\s*(.*?)\s*```"
         match = re.search(pattern, content, re.DOTALL)
         if match:
             try:
-                parsed = json.loads(match.group(1))
-                logger.info("Parsed JSON from TextMessage: %s", parsed)
-                return parsed
+                json_str = match.group(1).strip()
+                logger.debug("Found JSON in code block: %s", json_str)
+                return json.loads(json_str)
             except json.JSONDecodeError as e:
-                logger.error("Failed to parse JSON from TextMessage: %s, Content: %s", e, content)
-        # Fallback: Try raw JSON array
-        json_pattern = r"\[\s*\{.*?\}\s*\]"
-        match = re.search(json_pattern, content, re.DOTALL)
-        if match:
-            try:
-                parsed = json.loads(match.group(0))
-                logger.info("Parsed fallback JSON from TextMessage: %s", parsed)
-                return parsed
-            except json.JSONDecodeError as e:
-                logger.error("Failed to parse fallback JSON from TextMessage: %s, Content: %s", e, content)
-        # Fallback: Try any JSON-like structure
+                logger.error("Failed to parse JSON from code block: %s", e)
+        
+        # Try to find arrays or objects
+        json_patterns = [
+            r"\[\s*\{.*?\}\s*\]",  # Array of objects
+            r"\{\s*\".*?\"\s*:.*?\}",  # Object
+        ]
+        
+        for pattern in json_patterns:
+            match = re.search(pattern, content, re.DOTALL)
+            if match:
+                try:
+                    json_str = match.group(0).strip()
+                    logger.debug("Found JSON with pattern %s: %s", pattern, json_str)
+                    return json.loads(json_str)
+                except json.JSONDecodeError as e:
+                    logger.error("Failed to parse JSON with pattern %s: %s", pattern, e)
+        
+        # Try to find JSON anywhere in the content
         try:
-            parsed = json.loads(content)
-            if isinstance(parsed, (list, dict)):
-                logger.info("Parsed JSON from raw content: %s", parsed)
-                return parsed
-        except json.JSONDecodeError:
-            pass
-        logger.warning("No JSON found in TextMessage content: %s", content)
+            for i in range(len(content)):
+                for j in range(len(content), i, -1):
+                    substring = content[i:j].strip()
+                    if (substring.startswith('{') and substring.endswith('}')) or \
+                       (substring.startswith('[') and substring.endswith(']')):
+                        try:
+                            parsed = json.loads(substring)
+                            if isinstance(parsed, (list, dict)):
+                                logger.info("Found JSON in substring: %s", substring)
+                                return parsed
+                        except json.JSONDecodeError:
+                            continue
+        except Exception as e:
+            logger.error("Error in JSON substring search: %s", e)
+        
+        logger.warning("No JSON found in TextMessage content")
         return None
     
     elif isinstance(message, StructuredMessage):
@@ -257,69 +279,59 @@ def extract_json_from_message(message):
         for ctx_msg in message.context:
             if hasattr(ctx_msg, "content"):
                 content = ctx_msg.content
-                logger.debug("Handoff context message content: %s", content)
+                logger.debug("HandoffMessage context content: %s", content)
                 if isinstance(content, str):
                     pattern = r"```json\s*(.*?)\s*```"
                     match = re.search(pattern, content, re.DOTALL)
                     if match:
                         try:
-                            parsed = json.loads(match.group(1))
-                            logger.info("Parsed JSON from HandoffMessage context: %s", parsed)
-                            return parsed
-                        except json.JSONDecodeError as e:
-                            logger.error("Failed to parse JSON from HandoffMessage context: %s, Content: %s", e, content)
-                    json_pattern = r"\[\s*\{.*?\}\s*\]"
-                    match = re.search(json_pattern, content, re.DOTALL)
-                    if match:
-                        try:
-                            parsed = json.loads(match.group(0))
-                            logger.info("Parsed fallback JSON from HandoffMessage context: %s", parsed)
-                            return parsed
+                            return json.loads(match.group(1))
                         except json.JSONDecodeError as e:
-                            logger.error("Failed to parse fallback JSON from HandoffMessage context: %s, Content: %s", e, content)
-                    try:
-                        parsed = json.loads(content)
-                        if isinstance(parsed, (list, dict)):
-                            logger.info("Parsed JSON from raw HandoffMessage context: %s", parsed)
-                            return parsed
-                    except json.JSONDecodeError:
-                        pass
+                            logger.error("Failed to parse JSON from HandoffMessage: %s", e)
+                    
+                    json_patterns = [
+                        r"\[\s*\{.*?\}\s*\]",  # Array of objects
+                        r"\{\s*\".*?\"\s*:.*?\}",  # Object
+                    ]
+                    
+                    for pattern in json_patterns:
+                        match = re.search(pattern, content, re.DOTALL)
+                        if match:
+                            try:
+                                return json.loads(match.group(0))
+                            except json.JSONDecodeError as e:
+                                logger.error("Failed to parse JSON with pattern %s: %s", pattern, e)
                 elif isinstance(content, dict):
                     return content.get("slides", content)
+        
         logger.warning("No JSON found in HandoffMessage context")
         return None
     
     logger.warning("Unsupported message type for JSON extraction: %s", type(message))
     return None
 
-# Function to generate Markdown and convert to PDF (portrait, centered)
-def generate_slides_pdf(slides):
-    pdf = MarkdownPdf()
-    
-    for slide in slides:
-        content_lines = slide['content'].replace('\n', '\n\n')
-        markdown_content = f"""
-<div style="display: flex; flex-direction: column; justify-content: center; align-items: center; height: 100%; text-align: center; padding: 20px;">
-# {slide['title']}
-
-*Prof. AI Feynman*  
-*Princeton University, April 26th, 2025*
-
-{content_lines}
+# Function to generate HTML slides
+def generate_html_slides(slides, title):
+    try:
+        slides_html = ""
+        
+        for i, slide in enumerate(slides):
+            content_lines = slide['content'].replace('\n', '<br>')
+            slide_html = f"""
+<div id="slide-{i+1}" class="slide" style="display: none; height: 100%; padding: 20px; text-align: center;">
+    <h1 style="margin-bottom: 10px;">{slide['title']}</h1>
+    <h3 style="margin-bottom: 20px; font-style: italic;">Prof. AI Feynman<br>Princeton University, April 26th, 2025</h3>
+    <div style="font-size: 1.2em; line-height: 1.6;">{content_lines}</div>
 </div>
-
----
 """
-        pdf.add_section(Section(markdown_content, toc=False))
-    
-    pdf_file = os.path.join(OUTPUT_DIR, "slides.pdf")
-    try:
-        pdf.save(pdf_file)
-        logger.info("Generated PDF slides (portrait): %s", pdf_file)
-        return pdf_file
+            slides_html += slide_html
+        
+        logger.info(f"Generated HTML slides for: {title}")
+        return slides_html
     except Exception as e:
-        logger.error("Failed to generate PDF: %s", str(e))
-        raise
+        logger.error(f"Failed to generate HTML slides: {str(e)}")
+        logger.error(traceback.format_exc())
+        return None
 
 # Async function to update audio preview
 async def update_audio_preview(audio_file):
@@ -339,10 +351,6 @@ async def on_generate(api_service, api_key, serpapi_key, title, topic, instructi
         """
         return
 
-    # Ensure output directory exists
-    os.makedirs(OUTPUT_DIR, exist_ok=True)
-    logger.info("Output directory set to: %s", OUTPUT_DIR)
-
     # Initialize TTS model
     tts = None
     try:
@@ -362,6 +370,9 @@ async def on_generate(api_service, api_key, serpapi_key, title, topic, instructi
 
     model_client = get_model_client(api_service, api_key)
     
+    actual_content_slides = num_slides
+    total_slides = actual_content_slides + 3  # Content slides + quiz, assignment, thank-you
+    
     research_agent = AssistantAgent(
         name="research_agent",
         model_client=model_client,
@@ -374,8 +385,8 @@ async def on_generate(api_service, api_key, serpapi_key, title, topic, instructi
         model_client=model_client,
         handoffs=["script_agent"],
         system_message=f"""
-You are a Slide Agent. Using the research from the conversation history, generate EXACTLY {num_slides} content slides, plus 1 quiz slide, 1 assignment slide, and 1 thank-you slide, for a TOTAL of {num_slides + 3} slides. Output ONLY a JSON array wrapped in ```json ... ``` in a TextMessage, with each slide as an object with 'title' and 'content' keys. Ensure the JSON is valid, contains EXACTLY {num_slides + 3} slides, and matches the specified count before proceeding. Do not include explanatory text, comments, or other messages. After outputting, use the handoff_to_script_agent tool.
-Example for 2 content slides:
+You are a Slide Agent. Using the research from the conversation history and the specified number of content slides ({actual_content_slides}), generate exactly {actual_content_slides} content slides, plus one quiz slide, one assignment slide, and one thank-you slide, for a total of {total_slides} slides. Output ONLY a JSON array wrapped in ```json ... ``` in a TextMessage, where each slide is an object with 'title' and 'content' keys. Do not include any explanatory text, comments, or other messages. Ensure the JSON is valid and contains exactly {total_slides} slides before proceeding. After outputting the JSON, use the handoff_to_script_agent tool to pass the task to the Script Agent.
+Example output for 2 content slides:
 ```json
 [
     {{"title": "Slide 1", "content": "Content for slide 1"}},
@@ -393,11 +404,13 @@ Example for 2 content slides:
         model_client=model_client,
         handoffs=["feynman_agent"],
         system_message=f"""
-You are a Script Agent. Access the JSON array of {num_slides + 3} slides from the conversation history. Generate a narration script (1-2 sentences) for each of the {num_slides + 3} slides, summarizing its content in a natural, conversational tone as a speaker would, including occasional non-verbal words (e.g., "um," "you know," "like"). Output ONLY a JSON array wrapped in ```json ... ``` with exactly {num_slides + 3} strings, one script per slide, in the same order. Ensure the JSON is valid and complete. After outputting, use the handoff_to_feynman_agent tool. If scripts cannot be generated, retry once.
-Example for 1 content slide:
+You are a Script Agent. Access the JSON array of {total_slides} slides from the conversation history. Generate a narration script (1-2 sentences) for each of the {total_slides} slides, summarizing its content in a natural, conversational tone as a speaker would, including occasional non-verbal words (e.g., "um," "you know," "like"). Output ONLY a JSON array wrapped in ```json ... ``` with exactly {total_slides} strings, one script per slide, in the same order. Ensure the JSON is valid and complete. After outputting, use the handoff_to_feynman_agent tool. If scripts cannot be generated, retry once.
+Example for 3 content slides:
 ```json
 [
     "So, this slide, um, covers the main topic in a fun way.",
+    "The second slide introduces the key concepts.",
+    "This third slide shows some interesting applications.",
     "Alright, you know, answer these quiz questions.",
     "Here's your, like, assignment to complete.",
     "Thanks for, um, attending today!"
@@ -411,8 +424,8 @@ Example for 1 content slide:
         model_client=model_client,
         handoffs=[],
         system_message=f"""
-You are Agent Feynman. Review the slides and scripts from the conversation history to ensure coherence, completeness, and that EXACTLY {num_slides + 3} slides and {num_slides + 3} scripts are received. Output a confirmation message summarizing the number of slides and scripts received. If slides or scripts are missing, invalid, or do not match the expected count ({num_slides + 3}), report the issue clearly. Use 'TERMINATE' to signal completion.
-Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture is coherent. TERMINATE'
+You are Agent Feynman. Review the slides and scripts from the conversation history to ensure coherence, completeness, and that exactly {total_slides} slides and {total_slides} scripts are received. Output a confirmation message summarizing the number of slides and scripts received. If slides or scripts are missing, invalid, or do not match the expected count ({total_slides}), report the issue clearly. Use 'TERMINATE' to signal completion.
+Example: 'Received {total_slides} slides and {total_slides} scripts. Lecture is coherent. TERMINATE'
 """)
     
     swarm = Swarm(
@@ -430,10 +443,10 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
     Topic: {topic}
     Additional Instructions: {instructions}
     Audience: {lecture_type}
-    Number of Content Slides: {num_slides}
+    Number of Content Slides: {actual_content_slides}
     Please start by researching the topic.
     """
-    logger.info("Starting lecture generation for topic: %s", topic)
+    logger.info("Starting lecture generation for topic: %s with %d content slides", topic, actual_content_slides)
     
     slides = None
     scripts = None
@@ -445,144 +458,147 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
     """
     
     try:
-        max_slide_retries = 2
+        logger.info("Research Agent starting...")
+        task_result = await Console(swarm.run_stream(task=initial_message))
+        logger.info("Swarm execution completed")
+        
         slide_retry_count = 0
+        script_retry_count = 0
+        max_retries = 2
         
-        while slide_retry_count <= max_slide_retries:
-            logger.info("Slide generation attempt %d/%d", slide_retry_count + 1, max_slide_retries)
-            task_result = await Console(swarm.run_stream(task=initial_message))
-            logger.info("Swarm execution completed")
+        for message in task_result.messages:
+            source = getattr(message, 'source', getattr(message, 'sender', None))
+            logger.debug("Processing message from %s, type: %s", source, type(message))
             
-            script_retry_count = 0
-            max_script_retries = 2
-            
-            for message in task_result.messages:
-                source = getattr(message, 'source', getattr(message, 'sender', None))
-                logger.debug("Processing message from %s, type: %s, content: %s", source, type(message), message.to_text() if hasattr(message, 'to_text') else str(message))
-                
-                if isinstance(message, HandoffMessage):
-                    logger.info("Handoff from %s to %s", source, message.target)
-                    if source == "research_agent" and message.target == "slide_agent":
-                        progress = 25
-                        label = "Slides: generating..."
-                        yield html_with_progress(label, progress)
-                        await asyncio.sleep(0.1)
-                    elif source == "slide_agent" and message.target == "script_agent":
-                        if slides is None:
-                            logger.warning("Slide Agent handoff without slides JSON")
-                            extracted_json = extract_json_from_message(message)
-                            if extracted_json:
-                                slides = extracted_json
-                                logger.info("Extracted slides JSON from HandoffMessage context: %s", slides)
-                        if slides is None:
-                            label = "Slides: failed to generate..."
-                            yield html_with_progress(label, progress)
-                            await asyncio.sleep(0.1)
-                        progress = 50
-                        label = "Scripts: generating..."
-                        yield html_with_progress(label, progress)
-                        await asyncio.sleep(0.1)
-                    elif source == "script_agent" and message.target == "feynman_agent":
-                        if scripts is None:
-                            logger.warning("Script Agent handoff without scripts JSON")
-                            extracted_json = extract_json_from_message(message)
-                            if extracted_json:
-                                scripts = extracted_json
-                                logger.info("Extracted scripts JSON from HandoffMessage context: %s", scripts)
-                        progress = 75
-                        label = "Review: in progress..."
-                        yield html_with_progress(label, progress)
-                        await asyncio.sleep(0.1)
-                
-                elif source == "research_agent" and isinstance(message, TextMessage) and "handoff_to_slide_agent" in message.content:
-                    logger.info("Research Agent completed research")
+            if isinstance(message, HandoffMessage):
+                logger.info("Handoff from %s to %s", source, message.target)
+                if source == "research_agent" and message.target == "slide_agent":
                     progress = 25
                     label = "Slides: generating..."
                     yield html_with_progress(label, progress)
                     await asyncio.sleep(0.1)
-                
-                elif source == "slide_agent" and isinstance(message, (TextMessage, StructuredMessage)):
-                    logger.debug("Slide Agent message received: %s", message.to_text())
-                    extracted_json = extract_json_from_message(message)
-                    if extracted_json:
-                        slides = extracted_json
-                        logger.info("Slide Agent generated %d slides: %s", len(slides), slides)
-                        # Save slide content to individual files
-                        for i, slide in enumerate(slides):
-                            content_file = os.path.join(OUTPUT_DIR, f"slide_{i+1}_content.txt")
-                            try:
-                                with open(content_file, "w", encoding="utf-8") as f:
-                                    f.write(slide["content"])
-                                logger.info("Saved slide content to %s: %s", content_file, slide["content"])
-                            except Exception as e:
-                                logger.error("Error saving slide content to %s: %s", content_file, str(e))
-                        progress = 50
-                        label = "Scripts: generating..."
-                        yield html_with_progress(label, progress)
-                        await asyncio.sleep(0.1)
-                    else:
-                        logger.warning("No JSON extracted from slide_agent message: %s", message.to_text())
-                
-                elif source == "script_agent" and isinstance(message, (TextMessage, StructuredMessage)):
-                    logger.debug("Script Agent message received: %s", message.to_text())
-                    extracted_json = extract_json_from_message(message)
-                    if extracted_json:
-                        scripts = extracted_json
-                        logger.info("Script Agent generated scripts for %d slides: %s", len(scripts), scripts)
-                        # Save raw scripts to individual files
-                        for i, script in enumerate(scripts):
-                            script_file = os.path.join(OUTPUT_DIR, f"slide_{i+1}_raw_script.txt")
-                            try:
-                                with open(script_file, "w", encoding="utf-8") as f:
-                                    f.write(script)
-                                logger.info("Saved raw script to %s: %s", script_file, script)
-                            except Exception as e:
-                                logger.error("Error saving raw script to %s: %s", script_file, str(e))
-                        progress = 75
-                        label = "Scripts generated and saved. Reviewing..."
-                        yield html_with_progress(label, progress)
-                        await asyncio.sleep(0.1)
-                    else:
-                        logger.warning("No JSON extracted from script_agent message: %s", message.to_text())
-                        if script_retry_count < max_script_retries:
-                            script_retry_count += 1
-                            logger.info("Retrying script generation (attempt %d/%d)", script_retry_count, max_script_retries)
+                elif source == "slide_agent" and message.target == "script_agent":
+                    if slides is None:
+                        logger.warning("Slide Agent handoff without slides JSON")
+                        extracted_json = extract_json_from_message(message)
+                        if extracted_json:
+                            slides = extracted_json
+                            logger.info("Extracted slides JSON from HandoffMessage context: %s", slides)
+                    if slides is None or len(slides) != total_slides:
+                        if slide_retry_count < max_retries:
+                            slide_retry_count += 1
+                            logger.info("Retrying slide generation (attempt %d/%d)", slide_retry_count, max_retries)
                             retry_message = TextMessage(
-                                content="Please generate scripts for the slides as per your instructions.",
+                                content=f"Please generate exactly {total_slides} slides ({actual_content_slides} content slides plus quiz, assignment, thank-you) as per your instructions.",
                                 source="user",
-                                recipient="script_agent"
+                                recipient="slide_agent"
                             )
                             task_result.messages.append(retry_message)
                             continue
-                
-                elif source == "feynman_agent" and isinstance(message, TextMessage) and "TERMINATE" in message.content:
-                    logger.info("Feynman Agent completed lecture review: %s", message.content)
-                    progress = 90
-                    label = "Lecture materials ready. Generating audio..."
+                    progress = 50
+                    label = "Scripts: generating..."
                     yield html_with_progress(label, progress)
                     await asyncio.sleep(0.1)
+                elif source == "script_agent" and message.target == "feynman_agent":
+                    if scripts is None:
+                        logger.warning("Script Agent handoff without scripts JSON")
+                        extracted_json = extract_json_from_message(message)
+                        if extracted_json:
+                            scripts = extracted_json
+                            logger.info("Extracted scripts JSON from HandoffMessage context: %s", scripts)
+                    progress = 75
+                    label = "Review: in progress..."
+                    yield html_with_progress(label, progress)
+                    await asyncio.sleep(0.1)
+            
+            elif source == "research_agent" and isinstance(message, TextMessage) and "handoff_to_slide_agent" in message.content:
+                logger.info("Research Agent completed research")
+                progress = 25
+                label = "Slides: generating..."
+                yield html_with_progress(label, progress)
+                await asyncio.sleep(0.1)
             
-            # Validate slide count
-            expected_slide_count = num_slides + 3
-            if slides and len(slides) == expected_slide_count:
-                logger.info("Slide count validated: %d slides received", len(slides))
-                break
-            else:
-                logger.warning("Incorrect slide count: expected %d, got %d", expected_slide_count, len(slides) if slides else 0)
-                slide_retry_count += 1
-                slides = None
-                if slide_retry_count <= max_slide_retries:
-                    logger.info("Retrying slide generation (attempt %d/%d)", slide_retry_count + 1, max_slide_retries)
-                    task_result = await Console(swarm.run_stream(task=initial_message))
+            elif source == "slide_agent" and isinstance(message, (TextMessage, StructuredMessage)):
+                logger.debug("Slide Agent message received")
+                extracted_json = extract_json_from_message(message)
+                if extracted_json:
+                    slides = extracted_json
+                    logger.info("Slide Agent generated %d slides", len(slides))
+                    if len(slides) != total_slides:
+                        if slide_retry_count < max_retries:
+                            slide_retry_count += 1
+                            logger.info("Retrying slide generation (attempt %d/%d)", slide_retry_count, max_retries)
+                            retry_message = TextMessage(
+                                content=f"Please generate exactly {total_slides} slides ({actual_content_slides} content slides plus quiz, assignment, thank-you) as per your instructions.",
+                                source="user",
+                                recipient="slide_agent"
+                            )
+                            task_result.messages.append(retry_message)
+                            continue
+                    # Save slide content to individual files
+                    for i, slide in enumerate(slides):
+                        content_file = os.path.join(OUTPUT_DIR, f"slide_{i+1}_content.txt")
+                        try:
+                            with open(content_file, "w", encoding="utf-8") as f:
+                                f.write(slide["content"])
+                            logger.info("Saved slide content to %s", content_file)
+                        except Exception as e:
+                            logger.error("Error saving slide content to %s: %s", content_file, str(e))
+                    progress = 50
+                    label = "Scripts: generating..."
+                    yield html_with_progress(label, progress)
+                    await asyncio.sleep(0.1)
                 else:
-                    logger.error("Max slide retries reached")
-                    yield f"""
-                    <div style="display: flex; flex-direction: column; justify-content: center; align-items: center; height: 100%; min-height: 700px; padding: 20px; text-align: center; border: 1px solid #ddd; border-radius: 8px;">
-                        <h2 style="color: #d9534f;">Incorrect number of slides</h2>
-                        <p style="margin-top: 20px;">Expected {expected_slide_count} slides ({num_slides} content slides + quiz, assignment, thank-you), but generated {len(slides) if slides else 0}. Please try again with a different model.</p>
-                    </div>
-                    """
-                    return
+                    logger.warning("No JSON extracted from slide_agent message")
+                    if slide_retry_count < max_retries:
+                        slide_retry_count += 1
+                        logger.info("Retrying slide generation (attempt %d/%d)", slide_retry_count, max_retries)
+                        retry_message = TextMessage(
+                            content=f"Please generate exactly {total_slides} slides ({actual_content_slides} content slides plus quiz, assignment, thank-you) as per your instructions.",
+                            source="user",
+                            recipient="slide_agent"
+                        )
+                        task_result.messages.append(retry_message)
+                        continue
+            
+            elif source == "script_agent" and isinstance(message, (TextMessage, StructuredMessage)):
+                logger.debug("Script Agent message received")
+                extracted_json = extract_json_from_message(message)
+                if extracted_json:
+                    scripts = extracted_json
+                    logger.info("Script Agent generated scripts for %d slides", len(scripts))
+                    # Save raw scripts to individual files
+                    for i, script in enumerate(scripts):
+                        script_file = os.path.join(OUTPUT_DIR, f"slide_{i+1}_raw_script.txt")
+                        try:
+                            with open(script_file, "w", encoding="utf-8") as f:
+                                f.write(script)
+                            logger.info("Saved raw script to %s", script_file)
+                        except Exception as e:
+                            logger.error("Error saving raw script to %s: %s", script_file, str(e))
+                    progress = 75
+                    label = "Scripts generated and saved. Reviewing..."
+                    yield html_with_progress(label, progress)
+                    await asyncio.sleep(0.1)
+                else:
+                    logger.warning("No JSON extracted from script_agent message")
+                    if script_retry_count < max_retries:
+                        script_retry_count += 1
+                        logger.info("Retrying script generation (attempt %d/%d)", script_retry_count, max_retries)
+                        retry_message = TextMessage(
+                            content=f"Please generate exactly {total_slides} scripts for the {total_slides} slides as per your instructions.",
+                            source="user",
+                            recipient="script_agent"
+                        )
+                        task_result.messages.append(retry_message)
+                        continue
+            
+            elif source == "feynman_agent" and isinstance(message, TextMessage) and "TERMINATE" in message.content:
+                logger.info("Feynman Agent completed lecture review: %s", message.content)
+                progress = 90
+                label = "Lecture materials ready. Generating audio..."
+                yield html_with_progress(label, progress)
+                await asyncio.sleep(0.1)
         
         logger.info("Slides state: %s", "Generated" if slides else "None")
         logger.info("Scripts state: %s", "Generated" if scripts else "None")
@@ -594,7 +610,22 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
             for msg in task_result.messages:
                 source = getattr(msg, 'source', getattr(msg, 'sender', None))
                 logger.debug("Message from %s, type: %s, content: %s", source, type(msg), msg.to_text() if hasattr(msg, 'to_text') else str(msg))
-            yield error_html
+            yield f"""
+            <div style="display: flex; flex-direction: column; justify-content: center; align-items: center; height: 100%; min-height: 700px; padding: 20px; text-align: center; border: 1px solid #ddd; border-radius: 8px;">
+                <h2 style="color: #d9534f;">{error_message}</h2>
+                <p style="margin-top: 20px;">Please try again with a different model (e.g., Anthropic-claude-3-sonnet-20240229) or simplify the topic/instructions.</p>
+            </div>
+            """
+            return
+        
+        if len(slides) != total_slides:
+            logger.error("Expected %d slides (including %d content slides + 3), but received %d", total_slides, actual_content_slides, len(slides))
+            yield f"""
+            <div style="display: flex; flex-direction: column; justify-content: center; align-items: center; height: 100%; min-height: 700px; padding: 20px; text-align: center; border: 1px solid #ddd; border-radius: 8px;">
+                <h2 style="color: #d9534f;">Incorrect number of slides</h2>
+                <p style="margin-top: 20px;">Expected {total_slides} slides ({actual_content_slides} content slides + quiz, assignment, thank-you), but generated {len(slides)}. Please try again.</p>
+            </div>
+            """
             return
         
         if not isinstance(scripts, list) or not all(isinstance(s, str) for s in scripts):
@@ -607,7 +638,7 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
             """
             return
         
-        if len(scripts) != expected_slide_count:
+        if len(scripts) != total_slides:
             logger.error("Mismatch between number of slides (%d) and scripts (%d)", len(slides), len(scripts))
             yield f"""
             <div style="display: flex; flex-direction: column; justify-content: center; align-items: center; height: 100%; min-height: 700px; padding: 20px; text-align: center; border: 1px solid #ddd; border-radius: 8px;">
@@ -617,22 +648,19 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
             """
             return
         
-        # Generate PDF from slides
-        try:
-            pdf_file = generate_slides_pdf(slides)
-        except Exception as e:
-            logger.error("PDF generation failed: %s", str(e))
+        # Generate HTML slides
+        slides_html = generate_html_slides(slides, title)
+        if not slides_html:
+            logger.error("Failed to generate HTML slides")
             yield f"""
             <div style="display: flex; flex-direction: column; justify-content: center; align-items: center; height: 100%; min-height: 700px; padding: 20px; text-align: center; border: 1px solid #ddd; border-radius: 8px;">
-                <h2 style="color: #d9534f;">PDF generation failed</h2>
-                <p style="margin-top: 20px;">Error: {str(e)}</p>
-                <p>Please try again or check the lecture_generation.log for details.</p>
+                <h2 style="color: #d9534f;">Failed to generate slides</h2>
+                <p style="margin-top: 20px;">Please try again.</p>
             </div>
             """
             return
         
         audio_files = []
-        speaker_audio = speaker_audio if speaker_audio else "feynman.mp3"
         validated_speaker_wav = await validate_and_convert_speaker_audio(speaker_audio)
         if not validated_speaker_wav:
             logger.error("Invalid speaker audio after conversion, skipping TTS")
@@ -667,8 +695,8 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
                 await asyncio.sleep(0.1)
                 continue
             
-            max_retries = 2
-            for attempt in range(max_retries + 1):
+            max_audio_retries = 2
+            for attempt in range(max_audio_retries + 1):
                 try:
                     current_text = cleaned_script
                     if attempt > 0:
@@ -690,7 +718,7 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
                     break
                 except Exception as e:
                     logger.error("Error generating audio for slide %d (attempt %d): %s\n%s", i + 1, attempt, str(e), traceback.format_exc())
-                    if attempt == max_retries:
+                    if attempt == max_audio_retries:
                         logger.error("Max retries reached for slide %d, skipping", i + 1)
                         audio_files.append(None)
                         progress = 90 + ((i + 1) / len(scripts)) * 10
@@ -699,21 +727,15 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
                         await asyncio.sleep(0.1)
                         break
         
-        # Prepare output HTML with gr.File for PDF and gr.FileExplorer for outputs
         slides_info = json.dumps({"slides": [
             {"title": slide["title"], "content": slide["content"]}
             for slide in slides
         ], "audioFiles": audio_files})
         
         html_output = f"""
-        <div id="lecture-container" style="height: 700px; border: 1px solid #ddd; border-radius: 8px; display: flex; flex-direction: column; justify-content: space-between; padding: 20px;">
-            <div style="flex: 1; overflow: auto;">
-                <h3>Lecture Slides</h3>
-                <p>Download or view the slides PDF below (opens in your browser's PDF viewer):</p>
-                <gradio-file value="{pdf_file}" label="Slides PDF" file_types=[".pdf"]></gradio-file>
-                <h3>Generated Files</h3>
-                <p>Explore all generated files (PDF, audio, scripts) in the output directory:</p>
-                <gradio-file-explorer glob="/data/outputs/*" label="Output Directory"></gradio-file-explorer>
+        <div id="lecture-container" style="height: 700px; border: 1px solid #ddd; border-radius: 8px; display: flex; flex-direction: column; justify-content: space-between;">
+            <div id="slide-content" style="flex: 1; overflow: auto;">
+                {slides_html}
             </div>
             <div style="padding: 20px;">
                 <div id="progress-bar" style="width: 100%; height: 5px; background-color: #ddd; border-radius: 2px; margin-bottom: 10px;">
@@ -738,7 +760,7 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
 
             for (let i = 0; i < totalSlides; i++) {{
                 if (lectureData.audioFiles && lectureData.audioFiles[i]) {{
-                    const audio = new Audio('/gradio_api/file=' + lectureData.audioFiles[i]);
+                    const audio = new Audio(lectureData.audioFiles[i]);
                     audioElements.push(audio);
                 }} else {{
                     audioElements.push(null);
@@ -746,6 +768,9 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
             }}
 
             function updateSlide() {{
+                for (let i = 1; i <= totalSlides; i++) {{
+                    document.getElementById(`slide-${{i}}`).style.display = (i - 1 === currentSlide) ? 'block' : 'none';
+                }}
                 slideCounter.textContent = `Slide ${{currentSlide + 1}} of ${{totalSlides}}`;
                 progressFill.style.width = `${{(currentSlide + 1) / totalSlides * 100}}%`;
 
@@ -794,6 +819,9 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
                     }});
                 }}
             }});
+
+            // Initialize first slide
+            updateSlide();
         </script>
         """
         logger.info("Lecture generation completed successfully")
@@ -805,7 +833,7 @@ Example: 'Received {num_slides + 3} slides and {num_slides + 3} scripts. Lecture
         <div style="display: flex; flex-direction: column; justify-content: center; align-items: center; height: 100%; min-height: 700px; padding: 20px; text-align: center; border: 1px solid #ddd; border-radius: 8px;">
             <h2 style="color: #d9534f;">Error during lecture generation</h2>
             <p style="margin-top: 10px; font-size: 16px;">{str(e)}</p>
-            <p style="margin-top: 20px;">Please try again or check the lecture_generation.log for details.</p>
+            <p style="margin-top: 20px;">Please try again or adjust your inputs.</p>
         </div>
         """
         return