app.py

import sys
import os
import pandas as pd
import pdfplumber
import json
import gradio as gr
from typing import List
from concurrent.futures import ThreadPoolExecutor, as_completed
import hashlib
import shutil
import re
import psutil
import subprocess
import traceback
import torch
import copy
import time

# Configure environment variables and logging
os.environ["VLLM_LOGGING_LEVEL"] = "DEBUG"
if not torch.cuda.is_available():
    print("No GPU detected. Forcing CPU mode by setting CUDA_VISIBLE_DEVICES to an empty string.")
    os.environ["CUDA_VISIBLE_DEVICES"] = ""

persistent_dir = "/data/hf_cache"
os.makedirs(persistent_dir, exist_ok=True)

model_cache_dir = os.path.join(persistent_dir, "txagent_models")
tool_cache_dir = os.path.join(persistent_dir, "tool_cache")
file_cache_dir = os.path.join(persistent_dir, "cache")
report_dir = os.path.join(persistent_dir, "reports")
vllm_cache_dir = os.path.join(persistent_dir, "vllm_cache")

for directory in [model_cache_dir, tool_cache_dir, file_cache_dir, report_dir, vllm_cache_dir]:
    os.makedirs(directory, exist_ok=True)

os.environ["HF_HOME"] = model_cache_dir
os.environ["TRANSFORMERS_CACHE"] = model_cache_dir
os.environ["VLLM_CACHE_DIR"] = vllm_cache_dir
os.environ["TOKENIZERS_PARALLELISM"] = "false"
os.environ["CUDA_LAUNCH_BLOCKING"] = "1"

current_dir = os.path.dirname(os.path.abspath(__file__))
src_path = os.path.abspath(os.path.join(current_dir, "src"))
sys.path.insert(0, src_path)

from txagent.txagent import TxAgent

MEDICAL_KEYWORDS = {'diagnosis', 'assessment', 'plan', 'results', 'medications',
                    'allergies', 'summary', 'impression', 'findings', 'recommendations'}

def sanitize_utf8(text: str) -> str:
    return text.encode("utf-8", "ignore").decode("utf-8")

def file_hash(path: str) -> str:
    with open(path, "rb") as f:
        return hashlib.md5(f.read()).hexdigest()

def extract_priority_pages(file_path: str, max_pages: int = 20) -> str:
    try:
        text_chunks = []
        with pdfplumber.open(file_path) as pdf:
            # Always extract the first 3 pages
            for i, page in enumerate(pdf.pages[:3]):
                text = page.extract_text() or ""
                text_chunks.append(f"=== Page {i+1} ===\n{text.strip()}")
            # For pages 4 to max_pages, add only if medical keywords are found
            for i, page in enumerate(pdf.pages[3:max_pages], start=4):
                page_text = page.extract_text() or ""
                if any(re.search(rf'\b{kw}\b', page_text.lower()) for kw in MEDICAL_KEYWORDS):
                    text_chunks.append(f"=== Page {i} ===\n{page_text.strip()}")
        return "\n\n".join(text_chunks)
    except Exception as e:
        print("PDF processing error:", str(e))
        traceback.print_exc()
        return str(e)

def convert_file_to_json(file_path: str, file_type: str) -> str:
    try:
        h = file_hash(file_path)
        cache_path = os.path.join(file_cache_dir, f"{h}.json")
        if os.path.exists(cache_path):
            with open(cache_path, "r", encoding="utf-8") as f:
                return f.read()

        if file_type == "pdf":
            text = extract_priority_pages(file_path)
            result = json.dumps({"filename": os.path.basename(file_path), "content": text, "status": "initial"})
        elif file_type == "csv":
            df = pd.read_csv(file_path, encoding_errors="replace", header=None, dtype=str,
                             skip_blank_lines=False, on_bad_lines="skip")
            content = df.fillna("").astype(str).values.tolist()
            result = json.dumps({"filename": os.path.basename(file_path), "rows": content})
        elif file_type in ["xls", "xlsx"]:
            try:
                df = pd.read_excel(file_path, engine="openpyxl", header=None, dtype=str)
            except Exception:
                df = pd.read_excel(file_path, engine="xlrd", header=None, dtype=str)
            content = df.fillna("").astype(str).values.tolist()
            result = json.dumps({"filename": os.path.basename(file_path), "rows": content})
        else:
            result = json.dumps({"error": f"Unsupported file type: {file_type}"})

        with open(cache_path, "w", encoding="utf-8") as f:
            f.write(result)
        return result
    except Exception as e:
        print("Error processing", file_path, str(e))
        traceback.print_exc()
        return json.dumps({"error": str(e)})

def log_system_usage(tag=""):
    try:
        cpu = psutil.cpu_percent(interval=1)
        mem = psutil.virtual_memory()
        print(f"[{tag}] CPU: {cpu}% | RAM: {mem.used // (1024**2)}MB / {mem.total // (1024**2)}MB")
        result = subprocess.run(
            ["nvidia-smi", "--query-gpu=memory.used,memory.total,utilization.gpu", "--format=csv,nounits,noheader"],
            capture_output=True, text=True
        )
        if result.returncode == 0:
            used, total, util = result.stdout.strip().split(", ")
            print(f"[{tag}] GPU: {used}MB / {total}MB | Utilization: {util}%")
    except Exception as e:
        print(f"[{tag}] GPU/CPU monitor failed: {e}")
        traceback.print_exc()

def init_agent():
    try:
        print("🔁 Initializing model...")
        log_system_usage("Before Load")
        default_tool_path = os.path.abspath("data/new_tool.json")
        target_tool_path = os.path.join(tool_cache_dir, "new_tool.json")
        if not os.path.exists(target_tool_path):
            shutil.copy(default_tool_path, target_tool_path)

        agent = TxAgent(
            model_name="mims-harvard/TxAgent-T1-Llama-3.1-8B",
            rag_model_name="mims-harvard/ToolRAG-T1-GTE-Qwen2-1.5B",
            tool_files_dict={"new_tool": target_tool_path},
            force_finish=True,
            enable_checker=True,
            step_rag_num=8,
            seed=100,
            additional_default_tools=[],
        )
        agent.init_model()
        log_system_usage("After Load")
        print("✅ Agent Ready")
        return agent
    except Exception as e:
        print("❌ Error initializing agent:", str(e))
        traceback.print_exc()
        raise e

def create_ui(agent):
    with gr.Blocks(theme=gr.themes.Soft()) as demo:
        gr.Markdown("<h1 style='text-align: center;'>🩺 Clinical Oversight Assistant</h1>")
        # Persistent conversation state to maintain history
        conversation_state = gr.State([])
        chatbot = gr.Chatbot(label="Analysis", height=600, type="messages")
        file_upload = gr.File(file_types=[".pdf", ".csv", ".xls", ".xlsx"], file_count="multiple")
        msg_input = gr.Textbox(placeholder="Ask about potential oversights...", show_label=False)
        send_btn = gr.Button("Analyze", variant="primary")
        download_output = gr.File(label="Download Full Report")

        def analyze(message: str, state: list, files: list):
            if state is None:
                state = []
            history = state
            history.append({"role": "user", "content": message})
            history.append({"role": "assistant", "content": "⏳ Analyzing records for potential oversights..."})
            # Yield the initial update
            yield copy.deepcopy(history), None, copy.deepcopy(history)

            extracted = ""
            file_hash_value = ""
            if files:
                with ThreadPoolExecutor(max_workers=4) as executor:
                    futures = [executor.submit(convert_file_to_json, f.name, f.name.split(".")[-1].lower()) for f in files]
                    results = []
                    for future in as_completed(futures):
                        try:
                            res = future.result()
                            results.append(sanitize_utf8(res))
                        except Exception as e:
                            print("❌ Error in file processing:", str(e))
                            traceback.print_exc()
                extracted = "\n".join(results)
                file_hash_value = file_hash(files[0].name)

            prompt = f"""Review these medical records and identify EXACTLY what might have been missed:
1. List potential missed diagnoses
2. Flag any medication conflicts
3. Note incomplete assessments
4. Highlight abnormal results needing follow-up

Medical Records:
{extracted[:8000]}

### Potential Oversights:
"""
            print("🔎 Generated prompt:")
            print(prompt)

            full_response = ""
            response_chunks = []
            tool_calls_rendered = []

            try:
                for chunk in agent.run_gradio_chat(
                    message=prompt,
                    history=[],
                    temperature=0.2,
                    max_new_tokens=2048,
                    max_token=4096,
                    call_agent=False,
                    conversation=[]
                ):
                    if chunk is None:
                        continue
                    chunk_content = chunk if isinstance(chunk, str) else getattr(chunk, 'content', '')
                    if not chunk_content:
                        continue
                    response_chunks.append(chunk_content)
                    full_response = "".join(response_chunks)

                    # Collect and render any tool calls
                    matches = re.findall(r"\[TOOL_CALLS\]\[(.*?)\]", chunk_content, re.DOTALL)
                    for m in matches:
                        tool_calls_rendered.append(f"\n📦 Tool Call: [{m.strip()}]")

                    display_response = re.sub(r"\[TOOL_CALLS\].*?\n*", "", full_response, flags=re.DOTALL)
                    display_response = display_response.replace('[TxAgent]', '').strip()
                    display_response += "\n\n" + "\n".join(tool_calls_rendered)

                    if history and history[-1]["role"] == "assistant":
                        history[-1]["content"] = display_response
                    else:
                        history.append({"role": "assistant", "content": display_response})

                    # Yield updated conversation state
                    yield copy.deepcopy(history), None, copy.deepcopy(history)

                full_response = re.sub(r"\[TOOL_CALLS\].*?\n*", "", full_response, flags=re.DOTALL).strip()
                full_response = full_response.replace('[TxAgent]', '').strip()

                report_path = None
                if file_hash_value:
                    report_path = os.path.join(report_dir, f"{file_hash_value}_report.txt")
                    with open(report_path, "w", encoding="utf-8") as f:
                        f.write(full_response)

                if history and history[-1]["role"] == "assistant":
                    history[-1]["content"] = full_response
                else:
                    history.append({"role": "assistant", "content": full_response})

                yield copy.deepcopy(history), report_path if report_path and os.path.exists(report_path) else None, copy.deepcopy(history)

            except Exception as e:
                history.append({"role": "assistant", "content": f"❌ An error occurred in analyze: {str(e)}"})
                traceback.print_exc()
                yield copy.deepcopy(history), None, copy.deepcopy(history)

        send_btn.click(analyze, inputs=[msg_input, conversation_state, file_upload],
                       outputs=[chatbot, download_output, conversation_state])
        msg_input.submit(analyze, inputs=[msg_input, conversation_state, file_upload],
                         outputs=[chatbot, download_output, conversation_state])
    return demo

# Global variable to hold the WSGI/ASGI app for container environments.
app = None

if __name__ == "__main__":
    try:
        print("🚀 Launching app...")
        agent = init_agent()
        demo = create_ui(agent)
        # Launch the app with queueing; capture the returned app instance.
        launched_app, local_url, share_url = demo.queue(api_open=False).launch(
            server_name="0.0.0.0",
            server_port=7860,
            show_error=True,
            allowed_paths=[report_dir],
            share=False,
            ssr=False  # Disable SSR to improve UI updates
        )
        # Assign the underlying web app to the global variable for container access.
        app = launched_app.app
    except Exception as e:
        print("❌ Fatal error during launch:", str(e))
        traceback.print_exc()