app.py

# -*- coding: utf-8 -*-

import gradio as gr
import json
import aisuite as ai
import os
from dotenv import load_dotenv
load_dotenv()

# Setup client
os.environ["OPENAI_API_KEY"] = os.getenv("OPENAI_API_KEY")

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)