app.py

# ==========================================================
# PATHOGENAGENT - DEMO WITH MODULES 1 & 2
# ==========================================================

import gradio as gr
import json
import random
from transformers import AutoTokenizer, AutoModelForCausalLM

# ✅ Import ماژول‌ها
from modules.module_1_intent_router import IntentRouter
from modules.module_2_tool_executor import ToolExecutor

# ==========================================================
# ۱. بارگذاری دیتاست
# ==========================================================

def load_dataset():
    try:
        with open("biomedical_10k_dataset.json", "r", encoding="utf-8") as f:
            data = json.load(f)
        return data.get("questions", [])
    except:
        return []

QUESTIONS = load_dataset()
print(f"✅ {len(QUESTIONS)} questions loaded")

# ==========================================================
# ۲. راه‌اندازی ماژول‌ها و مدل
# ==========================================================

intent_router = IntentRouter()
tool_executor = ToolExecutor()

print("🔄 Loading BioGPT...")
MODEL_NAME = "Sepideh2027/biogpt-clinvar-finetuned"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
print("✅ Model loaded!")

# ==========================================================
# ۳. تابع اصلی
# ==========================================================

def get_random_question():
    if QUESTIONS:
        return random.choice(QUESTIONS).get("question", "")
    return "What is the clinical significance of CFTR F508del?"

def run_agent(query, use_random=False):
    if not query or use_random:
        query = get_random_question()
    
    # Step 1: Intent Detection (ماژول ۱)
    intent_result = intent_router.detect_intent(query)
    intent = intent_result["intent"]
    tools = intent_result["required_tools"]
    confidence = intent_result["confidence"]
    
    # Step 2: Evidence Retrieval (ماژول ۲)
    evidence = tool_executor.execute(query, tools)
    
    # Step 3: BioGPT Generation
    inputs = tokenizer(query, return_tensors="pt", truncation=True, max_length=512)
    outputs = model.generate(**inputs, max_new_tokens=100)
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    if query in response:
        response = response.split(query)[-1].strip()
    
    # Step 4: ساخت خروجی
    evidence_text = ""
    for source, items in evidence.items():
        if items:
            evidence_text += f"\n- **{source}:** " + ", ".join([i.get("title", "") for i in items[:3]])
        else:
            evidence_text += f"\n- **{source}:** No results"
    
    return f"""## 🧬 PathogenAgent

**Question:** {query}

**Intent:** {intent} (confidence: {confidence:.2f})  
**Tools Used:** {', '.join(tools)}

**Evidence:** {evidence_text}

**Answer:** {response}

---
*Powered by BioGPT + RAG*
"""

# ==========================================================
# ۴. رابط Gradio
# ==========================================================

with gr.Blocks(title="PathogenAgent", theme=gr.themes.Soft()) as demo:
    gr.Markdown("""
    # 🧬 PathogenAgent
    ### Evidence-Grounded AI Agent for Pathogen Genomics
    **BioGPT + RAG + PubMed/ClinVar/GenBank**
    """)
    
    with gr.Row():
        query_input = gr.Textbox(
            label="🔬 Enter your question",
            placeholder="e.g., What is the clinical significance of CFTR F508del?",
            lines=3,
            value="What is the clinical significance of CFTR F508del?"
        )
    
    with gr.Row():
        submit_btn = gr.Button("🚀 Run", variant="primary")
        random_btn = gr.Button("🎲 Random", variant="secondary")
    
    output = gr.Markdown(label="📝 Response")
    
    submit_btn.click(fn=run_agent, inputs=[query_input], outputs=[output])
    random_btn.click(fn=lambda: run_agent("", True), inputs=[], outputs=[output])

if __name__ == "__main__":
    demo.launch()