Upload 2 files

app.py

@@ -0,0 +1,363 @@
+# -*- coding: utf-8 -*-
+
+import gradio as gr
+import json
+import aisuite as ai
+import os
+from dotenv import load_dotenv
+load_dotenv()
+
+# Setup client
+client = ai.Client()
+
+def generate_draft(topic: str, model: str = "openai:gpt-4o") -> str:
+    prompt = f"""You are an expert essay writer with strong analytical skills.
+
+TASK: Write a compelling argumentative essay on the following topic:
+"{topic}"
+
+REQUIREMENTS:
+1. Structure: Introduction with clear thesis → 3 body paragraphs → Conclusion
+2. Each body paragraph should have: claim, evidence/reasoning, and connection to thesis
+3. Address at least one counterargument and refute it
+4. Use clear transitions between paragraphs
+5. Aim for 500 words
+
+Write the complete essay now:"""
+
+    response = client.chat.completions.create(
+        model=model,
+        messages=[{"role": "user", "content": prompt}],
+        temperature=1.0,
+    )
+    return response.choices[0].message.content
+
+
+def reflect_on_draft(draft: str, model: str = "openai:o4-mini") -> str:
+    prompt = f"""You are a rigorous writing instructor providing constructive feedback.
+
+ESSAY TO REVIEW:
+\"\"\"
+{draft}
+\"\"\"
+
+Analyze this essay across these dimensions and provide specific, actionable feedback:
+
+1. **THESIS CLARITY**: Is the main argument clear and specific?
+2. **ARGUMENT STRENGTH**: Are claims well-supported? Is reasoning logical?
+3. **EVIDENCE QUALITY**: Are examples concrete and relevant?
+4. **STRUCTURE**: Does the organization flow logically?
+5. **COUNTERARGUMENTS**: Are opposing views addressed fairly?
+6. **STYLE & CLARITY**: Is the writing concise and clear?
+
+For each dimension, identify what works well and what needs improvement.
+End with your TOP 3 PRIORITY improvements."""
+
+    response = client.chat.completions.create(
+        model=model,
+        messages=[{"role": "user", "content": prompt}],
+        temperature=1.0,
+    )
+    return response.choices[0].message.content
+
+
+def revise_draft(original_draft: str, reflection: str, model: str = "openai:gpt-4o") -> str:
+    prompt = f"""You are an expert editor tasked with improving an essay based on feedback.
+
+ORIGINAL ESSAY:
+\"\"\"
+{original_draft}
+\"\"\"
+
+FEEDBACK RECEIVED:
+\"\"\"
+{reflection}
+\"\"\"
+
+REVISION INSTRUCTIONS:
+1. Address EACH piece of feedback
+2. Strengthen the thesis if unclear
+3. Add concrete evidence where suggested
+4. Improve transitions between paragraphs
+5. Ensure counterarguments are properly addressed
+6. Fix any awkward phrasing
+
+IMPORTANT:
+- Return ONLY the complete revised essay
+- The revised essay MUST be at least 400 words
+- If the original is unclear, expand it into a full essay"""
+
+    response = client.chat.completions.create(
+        model=model,
+        messages=[{"role": "user", "content": prompt}],
+        temperature=1.0,
+    )
+    return response.choices[0].message.content
+
+
+# ============================================================
+# EVALUATION SYSTEM
+# ============================================================
+
+EVALUATION_CRITERIA = {
+    "thesis_clear": "The thesis is stated in ONE clear sentence",
+    "thesis_debatable": "The thesis makes a debatable claim",
+    "thesis_specific": "The thesis is narrow and focused",
+    "thesis_positioned": "The thesis appears in the introduction",
+    "multiple_reasons": "At least 3 DISTINCT supporting reasons",
+    "reasons_developed": "Each reason is explained in depth",
+    "logical_progression": "Arguments build on each other",
+    "no_logical_fallacies": "No logical fallacies present",
+    "concrete_examples": "Specific real-world examples included",
+    "named_sources": "References specific studies/stats/experts by name",
+    "evidence_explained": "Evidence is analyzed, not just dropped in",
+    "varied_evidence": "Multiple types of evidence used",
+    "counter_acknowledged": "At least one opposing view stated",
+    "counter_steelmanned": "Counterargument presented fairly",
+    "counter_refuted": "Substantive rebuttal provided",
+    "hook_present": "Introduction has an engaging hook",
+    "topic_sentences": "Each paragraph has a topic sentence",
+    "smooth_transitions": "Varied transitions between paragraphs",
+    "strong_conclusion": "Conclusion synthesizes (not just summarizes)",
+    "sentence_variety": "Varied sentence structures",
+    "precise_language": "Precise word choices (no vague terms)",
+    "no_repetition": "No excessive repetition",
+    "active_voice": "Predominantly active voice",
+    "no_filler": "No filler phrases",
+}
+
+
+def evaluate_essay(essay: str, model: str = "openai:gpt-4o") -> dict:
+    criteria_text = "\n".join([
+        f'{i+1}. "{name}": {desc}'
+        for i, (name, desc) in enumerate(EVALUATION_CRITERIA.items())
+    ])
+
+    prompt = f"""You are an EXTREMELY strict essay evaluator.
+
+ESSAY:
+\"\"\"
+{essay}
+\"\"\"
+
+CRITERIA (1 ONLY if FULLY met, otherwise 0):
+{criteria_text}
+
+RULES: Be HARSH. A typical draft should score 40-60%. If unsure, score 0.
+
+Respond in this JSON format:
+{{
+    "thesis_clear": 0 or 1,
+    "thesis_debatable": 0 or 1,
+    "thesis_specific": 0 or 1,
+    "thesis_positioned": 0 or 1,
+    "multiple_reasons": 0 or 1,
+    "reasons_developed": 0 or 1,
+    "logical_progression": 0 or 1,
+    "no_logical_fallacies": 0 or 1,
+    "concrete_examples": 0 or 1,
+    "named_sources": 0 or 1,
+    "evidence_explained": 0 or 1,
+    "varied_evidence": 0 or 1,
+    "counter_acknowledged": 0 or 1,
+    "counter_steelmanned": 0 or 1,
+    "counter_refuted": 0 or 1,
+    "hook_present": 0 or 1,
+    "topic_sentences": 0 or 1,
+    "smooth_transitions": 0 or 1,
+    "strong_conclusion": 0 or 1,
+    "sentence_variety": 0 or 1,
+    "precise_language": 0 or 1,
+    "no_repetition": 0 or 1,
+    "active_voice": 0 or 1,
+    "no_filler": 0 or 1
+}}
+
+Return ONLY valid JSON."""
+
+    response = client.chat.completions.create(
+        model=model,
+        messages=[{"role": "user", "content": prompt}],
+        temperature=0,
+    )
+
+    try:
+        result = json.loads(response.choices[0].message.content)
+    except json.JSONDecodeError:
+        content = response.choices[0].message.content
+        result = json.loads(content[content.find('{'):content.rfind('}')+1])
+
+    for key in result:
+        result[key] = 1 if result[key] else 0
+
+    result["total_score"] = sum(v for k, v in result.items() if k in EVALUATION_CRITERIA)
+    result["max_score"] = len(EVALUATION_CRITERIA)
+    result["percentage"] = round(100 * result["total_score"] / result["max_score"], 1)
+
+    return result
+
+
+def format_evaluation(eval_result: dict) -> str:
+    categories = {
+        "Thesis": ["thesis_clear", "thesis_debatable", "thesis_specific", "thesis_positioned"],
+        "Argument Depth": ["multiple_reasons", "reasons_developed", "logical_progression", "no_logical_fallacies"],
+        "Evidence Quality": ["concrete_examples", "named_sources", "evidence_explained", "varied_evidence"],
+        "Counterarguments": ["counter_acknowledged", "counter_steelmanned", "counter_refuted"],
+        "Structure": ["hook_present", "topic_sentences", "smooth_transitions", "strong_conclusion"],
+        "Writing Quality": ["sentence_variety", "precise_language", "no_repetition", "active_voice", "no_filler"],
+    }
+
+    lines = []
+    for cat, criteria in categories.items():
+        score = sum(eval_result[c] for c in criteria)
+        max_score = len(criteria)
+        checks = " ".join(["✅" if eval_result[c] else "❌" for c in criteria])
+        lines.append(f"{cat:<18} {checks}  ({score}/{max_score})")
+
+    lines.append("-" * 50)
+    lines.append(f"TOTAL: {eval_result['total_score']}/{eval_result['max_score']} ({eval_result['percentage']}%)")
+
+    return "\n".join(lines)
+
+
+def format_comparison(draft_eval: dict, revised_eval: dict) -> str:
+    categories = {
+        "Thesis": ["thesis_clear", "thesis_debatable", "thesis_specific", "thesis_positioned"],
+        "Argument Depth": ["multiple_reasons", "reasons_developed", "logical_progression", "no_logical_fallacies"],
+        "Evidence Quality": ["concrete_examples", "named_sources", "evidence_explained", "varied_evidence"],
+        "Counterarguments": ["counter_acknowledged", "counter_steelmanned", "counter_refuted"],
+        "Structure": ["hook_present", "topic_sentences", "smooth_transitions", "strong_conclusion"],
+        "Writing Quality": ["sentence_variety", "precise_language", "no_repetition", "active_voice", "no_filler"],
+    }
+
+    lines = []
+    lines.append(f"{'Category':<18} {'Draft':>10} {'Revised':>10} {'Change':>10}")
+    lines.append("=" * 50)
+
+    for cat, criteria in categories.items():
+        d = sum(draft_eval[c] for c in criteria)
+        r = sum(revised_eval[c] for c in criteria)
+        mx = len(criteria)
+        ch = r - d
+        ch_str = f"+{ch}" if ch > 0 else str(ch)
+        lines.append(f"{cat:<18} {d}/{mx}:>8 {r}/{mx}:>8 {ch_str:>10}")
+
+    lines.append("=" * 50)
+
+    # Fixed criteria
+    fixed = [c.replace("_", " ").title() for c in EVALUATION_CRITERIA
+             if draft_eval[c] == 0 and revised_eval[c] == 1]
+
+    imp = revised_eval["percentage"] - draft_eval["percentage"]
+    if imp > 0:
+        lines.append(f"\n📈 IMPROVEMENT: +{imp:.1f}%")
+        if fixed:
+            lines.append(f"\n✅ Fixed criteria:")
+            for c in fixed:
+                lines.append(f"   • {c}")
+
+    return "\n".join(lines)
+
+
+# ============================================================
+# MAIN WORKFLOW FUNCTION
+# ============================================================
+
+def run_reflection_workflow(essay_prompt: str, progress=gr.Progress()):
+    """Run the complete workflow and return all outputs."""
+
+    progress(0.1, desc="📝 Generating draft...")
+    draft = generate_draft(essay_prompt)
+
+    progress(0.3, desc="🧠 Reflecting on draft...")
+    feedback = reflect_on_draft(draft)
+
+    progress(0.5, desc="✍️ Revising draft...")
+    revised = revise_draft(draft, feedback)
+
+    progress(0.7, desc="📊 Evaluating draft...")
+    draft_eval = evaluate_essay(draft)
+
+    progress(0.85, desc="📊 Evaluating revision...")
+    revised_eval = evaluate_essay(revised)
+
+    progress(1.0, desc="✅ Complete!")
+
+    # Format outputs
+    draft_eval_text = format_evaluation(draft_eval)
+    revised_eval_text = format_evaluation(revised_eval)
+    comparison_text = format_comparison(draft_eval, revised_eval)
+
+    return draft, feedback, revised, draft_eval_text, revised_eval_text, comparison_text
+
+
+# ============================================================
+# GRADIO INTERFACE
+# ============================================================
+
+with gr.Blocks(title="Reflective Writing Agent") as demo:
+
+    gr.Markdown("""
+    # 🤖 Reflective Writing Agent
+    ### An Agentic AI Workflow: Draft → Reflect → Revise → Evaluate
+
+    Enter an essay prompt and watch the AI write, critique, and improve an essay —
+    with quantified evaluation showing exactly what improved.
+    """)
+
+    with gr.Row():
+        with gr.Column(scale=3):
+            prompt_input = gr.Textbox(
+                label="Essay Prompt",
+                placeholder="e.g., Should social media platforms be regulated by the government?",
+                lines=2
+            )
+        with gr.Column(scale=1):
+            run_btn = gr.Button("🚀 Run Workflow", variant="primary", size="lg")
+
+    gr.Markdown("---")
+
+    with gr.Tabs():
+        with gr.TabItem("📝 Step 1: Draft"):
+            draft_output = gr.Textbox(label="Initial Draft", lines=15)
+
+        with gr.TabItem("🧠 Step 2: Reflection"):
+            feedback_output = gr.Textbox(label="Feedback & Critique", lines=15)
+
+        with gr.TabItem("✍️ Step 3: Revision"):
+            revised_output = gr.Textbox(label="Revised Essay", lines=15)
+
+        with gr.TabItem("📊 Evaluation"):
+            with gr.Row():
+                with gr.Column():
+                    gr.Markdown("### Draft Evaluation")
+                    draft_eval_output = gr.Textbox(label="", lines=10)
+                with gr.Column():
+                    gr.Markdown("### Revised Evaluation")
+                    revised_eval_output = gr.Textbox(label="", lines=10)
+
+            gr.Markdown("### 📈 Comparison")
+            comparison_output = gr.Textbox(label="", lines=12)
+
+    # Connect button to function
+    run_btn.click(
+        fn=run_reflection_workflow,
+        inputs=[prompt_input],
+        outputs=[draft_output, feedback_output, revised_output,
+                 draft_eval_output, revised_eval_output, comparison_output]
+    )
+
+    # Example prompts
+    gr.Examples(
+        examples=[
+            ["Should social media platforms be regulated by the government?"],
+            ["Is artificial intelligence a threat to human employment?"],
+            ["Should college education be free for all students?"],
+            ["Are electric vehicles the solution to climate change?"],
+        ],
+        inputs=prompt_input
+    )
+
+# Launch the app
+if __name__ == "__main__":
+  demo.launch(theme=gr.themes.Soft(), share=True)

requirements.txt

@@ -0,0 +1,3 @@
+gradio>=6.0.0
+aisuite
+python-dotenv