Upload 2 files

app.py

@@ -1,8 +1,11 @@
 import gradio as gr
 import os
 import json
+import hashlib
 import shutil
 import time
+import re
+import anthropic
 from pathlib      import Path
 from dotenv       import load_dotenv
 from google       import genai
@@ -17,9 +20,43 @@ import io
 # On HF Spaces, set this in "Settings" -> "Secrets"
 load_dotenv() 
 API_KEY         = os.getenv("GOOGLE_API_KEY")
+CLAUDE_API_KEY  = os.getenv("CLAUDE_API_KEY")
 ACCESS_PASSWORD = os.getenv("APP_PASSWORD") 
 
-MODEL_NAME = "gemini-2.0-flash" 
+SCANNER_MODEL = "gemini-2.0-flash"
+COACH_MODEL   = "claude-sonnet-4-6"
+CACHE_DIR     = Path("cache/slides")
+CACHE_DIR.mkdir(parents=True, exist_ok=True)
+
+COACH_PERSONAS = {
+    "business": {
+        "name": "Business Strategy Coach",
+        "icon": "💼",
+        "role": "You are a Senior Business Strategist and executive communication expert.",
+        "focus": (
+            "Evaluate through a BUSINESS LENS:\n"
+            "- Is the business problem clearly articulated? Would a VP understand it?\n"
+            "- Does the executive summary lead with the answer, not the methodology?\n"
+            "- Is the value proposition compelling with specific ROI numbers?\n"
+            "- Is the business impact quantified and positioned persuasively?\n"
+            "- Would this presentation convince decision-makers to act?"
+        )
+    },
+    "analytics": {
+        "name": "Analytics & Methodology Coach",
+        "icon": "📊",
+        "role": "You are a Senior Data Scientist and ML methodology expert.",
+        "focus": (
+            "Evaluate through a TECHNICAL/ANALYTICAL LENS:\n"
+            "- Is the data structure and preparation approach well-documented?\n"
+            "- Are the target variables and evaluation metrics appropriate and justified?\n"
+            "- Is model selection rigorous? Were enough candidates explored?\n"
+            "- Is the HPO strategy systematic and well-explained?\n"
+            "- Is validation thorough (holdout tests, cross-validation, confidence intervals)?\n"
+            "- Are results reproducible from what is shown?"
+        )
+    }
+}
 
 # -----------------------------------------------------------------------------
 # LOGIC: CONVERSION (PDF -> IMAGES)
@@ -35,11 +72,11 @@ def convert_to_images(file_path):
     
     if ext == ".pdf":
         print("Converting PDF to images...")
-        images = convert_from_path(file_path, dpi=200) # 200 is fine for Gemini Vision
+        images = convert_from_path(file_path, dpi=150)
         image_paths = []
         for i, img in enumerate(images):
-            path = output_dir / f"slide-{i+1:02d}.png"
-            img.save(path, "PNG")
+            path = output_dir / f"slide-{i+1:02d}.jpg"
+            img.save(path, "JPEG", quality=85, optimize=True)
             image_paths.append(path)
         return image_paths
     else:
@@ -54,25 +91,37 @@ def scan_slides(client, image_paths):
     inventory = []
     total = len(image_paths)
     
-    # Process 1-by-1 to preserve context and allow progress updates
+    cache_hits = 0
+    
     for i, img_path in enumerate(image_paths):
         slide_num = i + 1
         yield f"Reading Slide {slide_num}/{total}...", None
         
+        with open(img_path, "rb") as f:
+            img_bytes = f.read()
+        
+        # Check slide cache by image hash
+        img_hash = hashlib.sha256(img_bytes).hexdigest()
+        cache_path = CACHE_DIR / f"{img_hash}.json"
+        
+        if cache_path.exists():
+            data = json.loads(cache_path.read_text())
+            data["slide_number"] = slide_num
+            inventory.append(data)
+            cache_hits += 1
+            print(f"   Slide {slide_num}: CACHE HIT")
+            continue
+        
         print(f"Scanning Slide {slide_num}...")
         
         # Rate Limiting: Sleep to respect API limits (avoid 429 errors)
-        # Check file size to adjust sleep time
-        file_size_mb = os.path.getsize(img_path) / (1024 * 1024)
+        file_size_mb = len(img_bytes) / (1024 * 1024)
         
         if file_size_mb > 1.0:
             print(f"   Large file ({file_size_mb:.1f}MB). Pausing 10s to refill quota...")
-            time.sleep(10) # Wait longer for big files
+            time.sleep(10)
         else:
-            time.sleep(2)  # Standard pause for normal files
-        
-        with open(img_path, "rb") as f:
-            img_bytes = f.read()
+            time.sleep(2)
 
         prompt = f"""
         Analyze this slide (Slide {slide_num}).
@@ -98,9 +147,9 @@ def scan_slides(client, image_paths):
         for attempt in range(max_retries):
             try:
                 response = client.models.generate_content(
-                    model=MODEL_NAME,
+                    model=SCANNER_MODEL,
                 contents=[
-                    types.Part.from_bytes(data=img_bytes, mime_type="image/png"),
+                    types.Part.from_bytes(data=img_bytes, mime_type="image/jpeg"),
                     prompt
                 ],
                     config=types.GenerateContentConfig(
@@ -110,7 +159,6 @@ def scan_slides(client, image_paths):
                 )
                 data = json.loads(response.text)
                 
-                # Robustness: Handle case where model returns a list [ { ... } ] instead of { ... }
                 if isinstance(data, list):
                     if len(data) > 0 and isinstance(data[0], dict):
                         data = data[0]
@@ -120,22 +168,24 @@ def scan_slides(client, image_paths):
 
                 if isinstance(data, dict):
                     inventory.append(data)
+                    cache_path.write_text(json.dumps(data, indent=2))
                 else:
                      print(f"Warning: Slide {slide_num} did not return a valid JSON dictionary. Data: {data}")
                 
-                # If successful, break retry loop
                 break
 
             except Exception as e:
                 error_str = str(e)
                 if "429" in error_str or "RESOURCE_EXHAUSTED" in error_str:
-                    wait_time = (attempt + 1) * 5 # 5, 10, 15, 20...
+                    wait_time = (attempt + 1) * 5
                     print(f"   ⚠️ Rate Limit (429) on Slide {slide_num}. Retrying in {wait_time}s...")
                     yield f"⚠️ High Traffic. Retrying Slide {slide_num} in {wait_time}s...", None
                     time.sleep(wait_time)
                 else:
                     print(f"Error scanning slide {slide_num}: {e}")
-                    break # Don't retry other errors
+                    break
+    
+    print(f"   Cache: {cache_hits}/{total} slides cached, {total - cache_hits} scanned via API")
             
     yield "Scan Complete", inventory
 
@@ -154,200 +204,217 @@ def debug_inventory(inventory):
 # -----------------------------------------------------------------------------
 # LOGIC: PASS 2 (COACH CRITIQUE)
 # -----------------------------------------------------------------------------
-def generate_critique(client, inventory, temperature=0.2):
-    try:
-        # Filter Appendix (Robustly)
-        def get_title(slide):
-            if not isinstance(slide, dict): return ""
-            t = slide.get("title")
-            return t if t else ""
-            
-        # Ensure 's' is a dict before accessing
-        active = [s for s in inventory if isinstance(s, dict) and "appendix" not in get_title(s).lower()]
-        print(f"DEBUG: Pass 2 Prompt contains {len(active)} active slides (excluding appendices).")
-        
-        # Script
-        script = []
-        for s in active:
-            visuals = s.get("visual_elements", {})
-            if not isinstance(visuals, dict): visuals = {} # Safety
-            
-            busy = "BUSY" if visuals.get("is_busy") else "OK"
-            title = s.get('title', 'No Title')
-            num = s.get('slide_number', '?')
-            takeaway = s.get('key_takeaway', '')
-            desc = s.get('visual_description', '')
-            
-            entry = f"Slide {num}: {title}\n- Content: {takeaway}\n- Visuals: {desc} [{busy}]"
-            script.append(entry)
-            
-        full_text = "\n".join(script)
-        
-        # THE MENTOR PROMPT (Synced with run_pass_2_gemini.py)
-        prompt = f"""
-        You are Dr. Jones, an expert Data Science Mentor. 
-        Your goal is to guide a student to professional excellence.
-        
-        SLIDE INVENTORY:
-        {full_text}
+def build_inventory_script(inventory):
+    """Shared logic: filter appendices and build the text script from inventory."""
+    def get_title(slide):
+        if not isinstance(slide, dict): return ""
+        t = slide.get("title")
+        return t if t else ""
+        
+    active = [s for s in inventory if isinstance(s, dict) and "appendix" not in get_title(s).lower()]
+    print(f"DEBUG: Pass 2 using {len(active)} active slides (excluding appendices).")
+    
+    script = []
+    for s in active:
+        visuals = s.get("visual_elements", {})
+        if not isinstance(visuals, dict): visuals = {}
         
-        TASK:
-        Coach this student based on the 8-Step Story Arc.
+        busy = "BUSY" if visuals.get("is_busy") else "OK"
+        title = s.get('title', 'No Title')
+        num = s.get('slide_number', '?')
+        takeaway = s.get('key_takeaway', '')
+        desc = s.get('visual_description', '')
         
-        REQUIRED STORY ARC:
-        1. Executive Summary
-        2. Data Structure
-        3. Targets & Metrics
-        4. Candidate Models
-        5. HPO Strategy
-        6. Best Model Selection
-        7. Validation
-        8. Business Impact
+        entry = f"Slide {num}: {title}\n- Content: {takeaway}\n- Visuals: {desc} [{busy}]"
+        script.append(entry)
         
-        INSTRUCTIONS:
-        1. **Fill the Roadmap**: For each of the 8 steps above, determine status (✅, ⚠️, ❓, ⭕).
-        2. **Check for Specifics**: If the student provides specific numbers (e.g. "$5,065 savings", "98% accuracy"), YOU MUST QUOTE THEM in the notes. Do not give generic advice if the specific data is present.
-        3. **Slide Refs**: Cite specific slide numbers in the notes.
-        4. **Tone**: Encouraging but precise.
-        5. **Summary**: Write a robust 2-paragraph summary (approx 150 words) explaining the overall impression and main areas for improvement.
+    return "\n".join(script)
+
+def generate_critique(coach_client, inventory, persona, temperature=0.2):
+    try:
+        full_text = build_inventory_script(inventory)
         
-        OUTPUT JSON SCHEMA:
-        {{
-            "overall_summary": "Encouraging feedback (2 paragraphs).",
-            "structure_roadmap": [
-                {{ 
-                    "step_name": "String (e.g. '1. Exec Summary')", 
-                    "status_icon": "String (✅, ⚠️, ❓, ⭕)", 
-                    "coach_notes": "String"
-                }}
-            ]
+        prompt = f"""{persona['role']}
+Your goal is to guide a Data Science student to professional excellence.
+
+{persona['focus']}
+
+SLIDE INVENTORY:
+{full_text}
+
+TASK:
+Coach this student based on the 8-Step Story Arc.
+
+REQUIRED STORY ARC:
+1. Executive Summary
+2. Data Structure
+3. Targets & Metrics
+4. Candidate Models
+5. HPO Strategy
+6. Best Model Selection
+7. Validation
+8. Business Impact
+
+INSTRUCTIONS:
+1. **Fill the Roadmap**: For each of the 8 steps above, determine status (✅, ⚠️, ❓, ⭕).
+2. **Check for Specifics**: If the student provides specific numbers (e.g. "$5,065 savings", "98% accuracy"), YOU MUST QUOTE THEM in the notes. Do not give generic advice if the specific data is present.
+3. **Slide Refs**: Cite specific slide numbers in the notes.
+4. **Tone**: Encouraging but precise.
+5. **Summary**: Write a robust 2-paragraph summary (approx 150 words) from your perspective as {persona['name']}.
+
+OUTPUT STRICT JSON (no markdown fences, no extra text):
+{{
+    "overall_summary": "Encouraging feedback (2 paragraphs).",
+    "structure_roadmap": [
+        {{ 
+            "step_name": "String (e.g. '1. Exec Summary')", 
+            "status_icon": "String (✅, ⚠️, ❓, ⭕)", 
+            "coach_notes": "String"
         }}
-        """
-        
-        response = client.models.generate_content(
-            model=MODEL_NAME,
-            contents=prompt,
-            config=types.GenerateContentConfig(
-                response_mime_type="application/json",
-                temperature=temperature
-            )
+    ]
+}}"""
+        
+        response = coach_client.messages.create(
+            model=COACH_MODEL,
+            max_tokens=4096,
+            temperature=temperature,
+            messages=[{"role": "user", "content": prompt}]
         )
-        print("DEBUG: Received Pass 2 Response from Gemini.")
         
-        critique = json.loads(response.text)
+        raw_text = response.content[0].text
+        print(f"DEBUG: {persona['name']} response received from {COACH_MODEL}.")
+        
+        cleaned = raw_text.strip()
+        fence_match = re.search(r"```(?:json)?\s*\n?(.*?)```", cleaned, re.DOTALL)
+        if fence_match:
+            cleaned = fence_match.group(1).strip()
+        
+        critique = json.loads(cleaned)
         
-        # Robustness: Handle list return from Coach
         if isinstance(critique, list):
             if len(critique) > 0 and isinstance(critique[0], dict):
                 critique = critique[0]
             else:
-                 # If it's a list but not a list of dicts, or empty, fail gracefully
                  raise ValueError(f"Coach returned a list, expected a dictionary. Output: {critique}")
                  
         return critique
         
     except Exception as e:
-        print(f"CRITICAL ERROR in Pass 2: {e}")
-        # Return a fallback critique so the UI doesn't hang
+        print(f"CRITICAL ERROR in Pass 2 ({persona['name']}): {e}")
         return {
             "overall_summary": f"Error generating critique: {e}",
             "structure_roadmap": [],
-            "slide_tips": []
         }
 
 # -----------------------------------------------------------------------------
 # GRADIO INTERFACE
 # -----------------------------------------------------------------------------
+def format_roadmap_table(critique):
+    """Build a markdown table from a critique's structure_roadmap."""
+    table_md = (
+        "| <span style='display:inline-block; min-width:180px'>STEP</span> "
+        "| <span style='display:inline-block; min-width:60px'>FLAG</span> "
+        "| COACH NOTES |\n|---|:---:|---|\n"
+    )
+    for item in critique.get("structure_roadmap", []):
+        icon = item.get('status_icon', '❓')
+        step = item.get('step_name', 'Step')
+        note = item.get('coach_notes', '')
+        table_md += f"| **{step}** | <span style='font-size: 1.5em'>{icon}</span> | {note} |\n"
+    return table_md
+
+EMPTY_OUTPUTS = ("", "", "", "", None, None, "")
+
 def process_presentation(file_obj, password, temperature):
     print("--- NEW JOB STARTED ---")
     if file_obj is None:
-        return "❌ Error: No file uploaded", "", "", None, None, ""
+        return ("❌ Error: No file uploaded",) + EMPTY_OUTPUTS
 
     if password != ACCESS_PASSWORD:
-        yield  "❌ Incorrect Password", "", "", None, None, ""
-        return "❌ Incorrect Password", "", "", None, None, ""
+        yield ("❌ Incorrect Password",) + EMPTY_OUTPUTS
+        return ("❌ Incorrect Password",) + EMPTY_OUTPUTS
         
     if not API_KEY:
-        return "❌ Server Error: API Key missing", "", "", None, None, ""
+        return ("❌ Server Error: Google API Key missing",) + EMPTY_OUTPUTS
+    if not CLAUDE_API_KEY:
+        return ("❌ Server Error: Claude API Key missing",) + EMPTY_OUTPUTS
         
-    client = genai.Client(api_key=API_KEY)
+    scanner_client = genai.Client(api_key=API_KEY)
+    coach_client   = anthropic.Anthropic(api_key=CLAUDE_API_KEY)
     
     try:
         # 1. Convert
         print("Step 1: Converting PDF...")
-        yield "Converting PDF...", "", "", None, None, "Converting PDF..."
+        yield ("Converting PDF...",) + EMPTY_OUTPUTS
         images = convert_to_images(file_obj.name)
         print(f"   Converted {len(images)} slides.")
         
-        # 2. Scan
-        yield f"Scanning {len(images)} slides... (Adaptive Speed)", \
-               "", "", None, None, "Starting Scan..."
+        # 2. Scan (Pass 1 - Gemini Flash)
+        yield (f"Scanning {len(images)} slides...",) + EMPTY_OUTPUTS
         print("Step 2: Scanning Slides (Pass 1)...")
         
-        # Iterate through the generator from scan_slides
-        scanner = scan_slides(client, images)
+        scanner = scan_slides(scanner_client, images)
         inventory = []
         
         for msg, result in scanner:
             if result is None:
-                # Still scanning, update status
-                yield msg, "", "", None, None, msg
+                yield (msg,) + EMPTY_OUTPUTS
             else:
-                # Scan complete, result is the inventory
                 inventory = result
                 
         print("   Scan Complete.")
         
-        # Save Inventory JSON (Artifact 1)
+        # Save Inventory
         original_stem = Path(file_obj.name).stem
-        
-        # Save to 'slides_images/[Stem]' to match project structure
         target_dir = Path("slides_images") / original_stem
         target_dir.mkdir(parents=True, exist_ok=True)
         
         inventory_filename = target_dir / f"{original_stem}_Inventory.json"
-        
         with open(inventory_filename, "w") as f:
             json.dump(inventory, f, indent=4)
         print(f"   Saved Inventory to {inventory_filename}")
         
-        # 3. Coach
-        debug_inventory(inventory) # Print debug info to console
+        # 3. Coach (Pass 2 - Sonnet 4.6, two personas)
+        debug_inventory(inventory)
+        
+        biz_persona = COACH_PERSONAS["business"]
+        ana_persona = COACH_PERSONAS["analytics"]
         
-        yield "Reviewing Presentation (Pass 2)...", "", "", \
-              None, None, "Analyzing Structure ..."
-        print(f"Step 3: Calling Coach (Pass 2) [Temp: {temperature}]...")
-        critique = generate_critique(client, inventory, temperature)
-        print("   Critique Received.")
+        yield (f"💼 {biz_persona['name']} reviewing...",) + EMPTY_OUTPUTS
+        print(f"Step 3a: {biz_persona['name']} [Temp: {temperature}]...")
+        biz_critique = generate_critique(coach_client, inventory, biz_persona, temperature)
+        print(f"   {biz_persona['name']} done.")
+        
+        yield (f"📊 {ana_persona['name']} reviewing...",) + EMPTY_OUTPUTS
+        print(f"Step 3b: {ana_persona['name']} [Temp: {temperature}]...")
+        ana_critique = generate_critique(coach_client, inventory, ana_persona, temperature)
+        print(f"   {ana_persona['name']} done.")
         
         # 4. Format Output
-        summary = critique.get("overall_summary", "")
-        
-        # Build Table Markdown
-        table_md = "| STEP| FLAG | COACH NOTES |\n|---|:---:|---|\n"
-        for item in critique.get("structure_roadmap", []): # Updated key to match schema
-            icon = item.get('status_icon', '❓')
-            step = item.get('step_name', 'Step')
-            note = item.get('coach_notes', '')
-            table_md += f"| **{step}** | <span style='font-size: 1.5em'>{icon}</span> | {note} |\n"
+        biz_summary  = biz_critique.get("overall_summary", "")
+        biz_table    = format_roadmap_table(biz_critique)
+        ana_summary  = ana_critique.get("overall_summary", "")
+        ana_table    = format_roadmap_table(ana_critique)
             
-        # Create Report File
-        # original_stem is already defined above
+        # Create Combined Report File
         report_filename = f"{original_stem}_Review.md"
-        
         with open(report_filename, "w") as f:
             f.write(f"# Dr. Jones Feedback for {original_stem}\n\n")
-            f.write("## Coach Summary\n")
-            f.write(summary + "\n\n")
-            f.write("## Story Roadmap\n")
-            f.write(table_md)
-            
-        # Return: Status, Summary, Table, Preview Image, Report File, Small Status
-        yield "✅ Done!", summary, table_md, images[0], report_filename, ""
+            f.write(f"## {biz_persona['icon']} {biz_persona['name']}\n\n")
+            f.write(biz_summary + "\n\n")
+            f.write("### Business Roadmap\n")
+            f.write(biz_table + "\n\n")
+            f.write(f"## {ana_persona['icon']} {ana_persona['name']}\n\n")
+            f.write(ana_summary + "\n\n")
+            f.write("### Analytics Roadmap\n")
+            f.write(ana_table)
+
+        yield "✅ Done!", biz_summary, biz_table, ana_summary, ana_table, \
+              images[0], report_filename, ""
         
     except Exception as e:
         print(f"CRITICAL ERROR: {e}")
-        yield f"❌ Error: {str(e)}", "", "", None, None, "Error"
+        yield (f"❌ Error: {str(e)}",) + EMPTY_OUTPUTS
 
 # Define a custom maroon color palette
 maroon = gr.themes.Color(
@@ -367,13 +434,13 @@ maroon = gr.themes.Color(
 with gr.Blocks(title="Dr. Jones AI Coach", 
         theme=gr.themes.Default(primary_hue=maroon, text_size="lg")) as demo:
     gr.Markdown("# 🎓 Capstone Slide Review")
-    gr.Markdown("Upload your slides (PDF) for feedback on your Capstone Story Arc.")
+    gr.Markdown("Upload your slides (PDF) for feedback from your AI coaching committee.")
     
     with gr.Row():
         with gr.Column(scale=3):
             file_input = gr.File(label="Upload PDF Slides", 
-                                 file_types=[".pdf"], type="filepath", 
-                                 height=150)
+                                 file_types=[".pdf", "application/pdf"], 
+                                 type="filepath", height=150)
         with gr.Column(scale=1):
             pass_input = gr.Textbox(label="Password", type="password")
             temp_input = gr.Slider(minimum=0.0, maximum=1.0, value=0.2, step=0.1, label="Coach Temperature")
@@ -382,28 +449,35 @@ with gr.Blocks(title="Dr. Jones AI Coach",
     
     status = gr.Markdown("**Status**: Ready")
     
-    # Results Area
     with gr.Row():
         with gr.Column(scale=1):
             preview_img = gr.Image(label="Title Slide", interactive=False)
-            with gr.Row():
-                download_btn = gr.File(label="Click Filesize to Download Report", scale=5)
-                with gr.Column(scale=1):
-                    progress_status = gr.Markdown(value="") # Temporary status next to download
-        
-        with gr.Column(scale=2):
-            gr.Markdown("### 👨‍🏫 Coach Summary")
-            summary_display = gr.Textbox(label="", show_label=False, lines=7, interactive=False)
+            download_btn = gr.File(label="Download Full Report")
+            progress_status = gr.Markdown(value="")
     
-    with gr.Group():
-        gr.Markdown("### 📋 The Story Roadmap")
-        roadmap_display = gr.Markdown()
+        with gr.Column(scale=2):
+            with gr.Tabs():
+                with gr.TabItem("💼 Business Strategy Coach"):
+                    biz_summary_display = gr.Textbox(label="Business Summary", 
+                                                     show_label=False, lines=6, interactive=False)
+                    
+                with gr.TabItem("📊 Analytics & Methodology Coach"):
+                    ana_summary_display = gr.Textbox(label="Analytics Summary", 
+                                                     show_label=False, lines=6, interactive=False)
+
+    with gr.Tabs():
+        with gr.TabItem("💼 Business Roadmap"):
+            biz_roadmap_display = gr.Markdown()
+            
+        with gr.TabItem("📊 Analytics Roadmap"):
+            ana_roadmap_display = gr.Markdown()
 
     btn.click(
         fn=process_presentation,
         inputs=[file_input, pass_input, temp_input],
-        outputs=[status, summary_display, roadmap_display, preview_img, 
-                 download_btn, progress_status]
+        outputs=[status, biz_summary_display, biz_roadmap_display,
+                 ana_summary_display, ana_roadmap_display,
+                 preview_img, download_btn, progress_status]
     )
 
 if __name__ == "__main__":

requirements.txt

@@ -1,5 +1,6 @@
 gradio
 google-genai
+anthropic
 python-dotenv
 pdf2image
 pillow