Create app.py

app.py

@@ -0,0 +1,217 @@
+import os
+from dotenv import load_dotenv
+import gradio as gr
+import google.generativeai as genai
+from PIL import Image
+import json
+
+load_dotenv()
+# 2. Configure the AI Model
+def setup_api_key():
+    try:
+        api_key = os.environ.get('GEMINI_API_KEY')
+        genai.configure(api_key=api_key)
+        print("🔑 API Key configured successfully.")
+        return True
+    except Exception as e:
+        print(f"🔴 Error during API configuration: {e}")
+        return False
+
+API_KEY_IS_SET = setup_api_key()
+
+# FIXED: Use correct, publicly available model names for stability.
+MODEL_VISION = 'gemini-1.5-flash-latest'
+MODEL_INITIAL = 'gemini-1.5-flash-latest'
+MODEL_REFINER = 'gemini-1.5-pro-latest'
+MODEL_REWRITER = 'gemini-1.5-pro-latest'
+
+# 3. Define System Instructions
+VISION_SYSTEM_INSTRUCTION = "You are an expert UI and content analyst..."
+PROMPT_ENGINEER_SYSTEM_INSTRUCTION = "You are a meticulous AI Prompt Engineer..."
+PROMPT_REFINER_SYSTEM_INSTRUCTION = "## CRITICAL SYSTEM DIRECTIVE\nYour output is not for a human..."
+META_PROMPT_SYSTEM_INSTRUCTION = """
+You are an AI Prompt Optimization expert, a 'Prompt Rewriter'. Your goal is to take a single input prompt and generate 3 distinct, improved variations.
+## CRITICAL OUTPUT FORMAT
+Your output MUST be a valid JSON array containing exactly three strings. Each string is a complete, rewritten prompt.
+"""
+
+# 4. Define AI processing functions
+def analyze_screenshot(pil_image: Image.Image) -> str:
+    model = genai.GenerativeModel(MODEL_VISION, system_instruction=VISION_SYSTEM_INSTRUCTION)
+    try: return model.generate_content(pil_image).text
+    except Exception as e: return f"Error: {e}"
+
+def initial_prompt_stream(analysis_text, goal):
+    model = genai.GenerativeModel(MODEL_INITIAL, system_instruction=PROMPT_ENGINEER_SYSTEM_INSTRUCTION)
+    prompt = f"**Analysis:**\n{analysis_text}\n\n**Goal:**\n{goal or 'Infer from analysis.'}"
+    for chunk in model.generate_content(prompt, stream=True): yield chunk.text
+
+def refinement_prompt_stream(original_prompt, feedback):
+    model = genai.GenerativeModel(MODEL_REFINER, system_instruction=PROMPT_REFINER_SYSTEM_INSTRUCTION)
+    prompt = f"**Original Prompt:**\n{original_prompt}\n\n**Feedback:**\n{feedback}"
+    for chunk in model.generate_content(prompt, stream=True): yield chunk.text
+
+def rewrite_prompt_with_prewrite(original_prompt):
+    model = genai.GenerativeModel(MODEL_REWRITER, system_instruction=META_PROMPT_SYSTEM_INSTRUCTION)
+    try:
+        response = model.generate_content(original_prompt)
+        cleaned_response = response.text.strip().replace("```json", "").replace("```", "")
+        variations = json.loads(cleaned_response)
+        if isinstance(variations, list) and len(variations) > 0:
+            return variations
+        return ["Error: AI returned an invalid format.", "", ""]
+    except Exception as e:
+        return [f"Error: {e}", "The AI's response was not valid JSON.", ""]
+
+# 5. Define Gradio App's interactive logic (CORRECTED)
+def generate_initial_prompt(pil_image, situation_text, goal_clarification):
+    # Reset UI
+    yield {
+        analysis_accordion: gr.update(visible=False, open=False, value=""),
+        final_prompt_output: gr.update(value=""),
+        satisfaction_row: gr.update(visible=False),
+        feedback_col: gr.update(visible=False),
+        prewrite_col: gr.update(visible=False),
+        prewrite_choices: gr.update(choices=[], value=None),
+        analysis_state: None,
+        first_prompt_state: None
+    }
+    if not API_KEY_IS_SET:
+        yield {analysis_accordion: gr.update(visible=True, open=True, value="API Key not set.")}; return
+    if pil_image is None and not situation_text.strip():
+        yield {analysis_accordion: gr.update(visible=True, open=True, value="Error: Please provide context.")}; return
+
+    yield {analysis_accordion: gr.update(visible=True, open=True, value="Processing...")}
+    analysis_text = analyze_screenshot(pil_image) if pil_image is not None else f"The user described their situation as: '{situation_text}'"
+    if "Error:" in analysis_text:
+        yield {analysis_accordion: gr.update(visible=True, open=True, value=analysis_text)}; return
+
+    yield {analysis_accordion: gr.update(value=analysis_text), final_prompt_output: gr.update(value="Generating...")}
+    final_prompt_full = ""
+    for chunk in initial_prompt_stream(analysis_text, goal_clarification):
+        final_prompt_full += chunk
+        # FIXED: THIS IS THE CRITICAL FIX.
+        # Every streaming update for the prompt MUST also contain the state of the analysis panel.
+        yield {
+            analysis_accordion: gr.update(visible=True, open=True, value=analysis_text),
+            final_prompt_output: gr.update(value=final_prompt_full.strip())
+        }
+    
+    # Final state update after stream is complete
+    yield {
+        satisfaction_row: gr.update(visible=True),
+        analysis_state: analysis_text,
+        first_prompt_state: final_prompt_full.strip()
+    }
+
+def handle_auto_refine(original_prompt):
+    variations = rewrite_prompt_with_prewrite(original_prompt)
+    return {
+        prewrite_col: gr.update(visible=True),
+        prewrite_choices: gr.update(choices=variations, value=variations[0] if variations else ""),
+        satisfaction_row: gr.update(visible=False),
+        feedback_col: gr.update(visible=False)
+    }
+
+def select_rewritten_prompt(selected_prompt):
+    return {
+        final_prompt_output: gr.update(value=selected_prompt),
+        first_prompt_state: selected_prompt,
+        satisfaction_row: gr.update(visible=True),
+        prewrite_col: gr.update(visible=False)
+    }
+    
+def handle_manual_feedback():
+    return {
+        feedback_col: gr.update(visible=True),
+        satisfaction_row: gr.update(visible=False),
+        prewrite_col: gr.update(visible=False)
+    }
+
+def handle_like():
+    return {
+        satisfaction_row: gr.update(visible=False),
+        feedback_col: gr.update(visible=False),
+        prewrite_col: gr.update(visible=False)
+    }
+
+def refine_with_manual_feedback(original_prompt, feedback, current_analysis):
+    if not feedback.strip():
+        yield {final_prompt_output: original_prompt}; return
+    yield {final_prompt_output: "Refining manually..."}
+    final_prompt_full = ""
+    for chunk in refinement_prompt_stream(original_prompt, feedback):
+        final_prompt_full += chunk
+        yield {
+            analysis_accordion: gr.update(visible=True, open=True, value=current_analysis),
+            final_prompt_output: final_prompt_full.strip(),
+            first_prompt_state: final_prompt_full.strip()
+        }
+    yield {satisfaction_row: gr.update(visible=True)}
+    
+def clear_all():
+    return {
+        image_input: None,
+        situation_input: "",
+        goal_input: "",
+        analysis_accordion: gr.update(visible=False, open=False, value=""),
+        final_prompt_output: "",
+        satisfaction_row: gr.update(visible=False),
+        feedback_col: gr.update(visible=False),
+        prewrite_col: gr.update(visible=False),
+        prewrite_choices: gr.update(choices=[], value=None),
+        analysis_state: None,
+        first_prompt_state: None
+    }
+
+# 6. Build and launch the Gradio Interface
+theme = gr.themes.Base(primary_hue=gr.themes.colors.blue, secondary_hue=gr.themes.colors.neutral, font=(gr.themes.GoogleFont("Inter"), "ui-sans-serif", "system-ui", "sans-serif")).set(button_primary_background_fill_hover='*primary_600', button_primary_border_color_hover='*primary_600')
+
+with gr.Blocks(theme=theme, analytics_enabled=False) as demo:
+    analysis_state, first_prompt_state = gr.State(), gr.State()
+    gr.Markdown("# 🤖 Context-Aware Prompt Engineer AI (v20 - FINAL)")
+    
+    with gr.Row(variant="panel"):
+        with gr.Column(scale=1):
+            with gr.Group():
+                gr.Markdown("### 1. Provide Context"); image_input = gr.Image(type="pil", label="Upload Screenshot"); gr.HTML("<div style='text-align: center; font-weight: 500; margin: 10px 0;'>OR</div>"); situation_input = gr.Textbox(label="Describe Your Situation", lines=6)
+            with gr.Accordion("2. (Optional) Clarify Goal", open=False):
+                goal_input = gr.Textbox(label=None, show_label=False)
+            with gr.Row():
+                submit_btn = gr.Button("🚀 Generate Prompt", variant="primary", scale=3)
+                clear_btn = gr.Button("🔄 Start Over", scale=1)
+        with gr.Column(scale=2):
+            with gr.Accordion("Show AI's Context Analysis", open=False) as analysis_accordion:
+                analysis_output = gr.Textbox(label=None, lines=7, interactive=False, show_label=False)
+            final_prompt_output = gr.Textbox(label="✅ Final Engineered Prompt", lines=12, interactive=False, show_copy_button=True)
+
+    with gr.Row(visible=False) as satisfaction_row:
+        gr.Markdown("#### Are you satisfied, or would you like to refine the prompt?"); 
+        like_btn = gr.Button("👍 Yes, Looks Good")
+        auto_refine_btn = gr.Button("🤖 Auto-Refine (PRewrite)")
+        dislike_btn = gr.Button("👎 Manual Feedback")
+
+    with gr.Column(visible=False) as prewrite_col:
+        gr.Markdown("### PRewrite: AI-Generated Variations"); prewrite_choices = gr.Radio(label="Select the best variation:", type="value"); select_version_btn = gr.Button("✅ Use This Version", variant="primary")
+
+    with gr.Column(visible=False) as feedback_col:
+        feedback_input = gr.Textbox(label="What would you like to change or improve?", lines=3); refine_btn = gr.Button("🛠️ Refine Manually", variant="primary")
+
+    # Event Handlers
+    outputs_for_generation = [analysis_accordion, final_prompt_output, satisfaction_row, feedback_col, prewrite_col, prewrite_choices, analysis_state, first_prompt_state]
+    submit_btn.click(fn=generate_initial_prompt, inputs=[image_input, situation_input, goal_input], outputs=outputs_for_generation)
+    goal_input.submit(fn=generate_initial_prompt, inputs=[image_input, situation_input, goal_input], outputs=outputs_for_generation)
+    
+    like_btn.click(fn=handle_like, outputs=[satisfaction_row, feedback_col, prewrite_col])
+    auto_refine_btn.click(fn=handle_auto_refine, inputs=[first_prompt_state], outputs=[prewrite_col, prewrite_choices, satisfaction_row, feedback_col])
+    dislike_btn.click(fn=handle_manual_feedback, outputs=[feedback_col, satisfaction_row, prewrite_col])
+    
+    select_version_btn.click(fn=select_rewritten_prompt, inputs=[prewrite_choices], outputs=[final_prompt_output, first_prompt_state, satisfaction_row, prewrite_col])
+    # Pass the analysis_state to the manual refiner
+    refine_btn.click(fn=refine_with_manual_feedback, inputs=[first_prompt_state, feedback_input, analysis_state], outputs=[analysis_accordion, final_prompt_output, first_prompt_state, satisfaction_row])
+    feedback_input.submit(fn=refine_with_manual_feedback, inputs=[first_prompt_state, feedback_input, analysis_state], outputs=[analysis_accordion, final_prompt_output, first_prompt_state, satisfaction_row])
+    
+    clear_btn.click(fn=clear_all, outputs=[image_input, situation_input, goal_input, analysis_accordion, final_prompt_output, satisfaction_row, feedback_col, prewrite_col, prewrite_choices, analysis_state, first_prompt_state])
+
+# FIXED: Removed invalid 'pwa' argument
+demo.launch(debug=True, pwa=True)