app.py

import sys
import os
import pandas as pd
import pdfplumber
import json
import gradio as gr
from typing import List, Tuple, Optional
from concurrent.futures import ThreadPoolExecutor, as_completed
import hashlib
import shutil
import re
import psutil
import subprocess
from datetime import datetime
import tiktoken

# Persistent directory setup
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

# Constants
MEDICAL_KEYWORDS = {
    'diagnosis', 'assessment', 'plan', 'results', 'medications',
    'allergies', 'summary', 'impression', 'findings', 'recommendations',
    'conclusion', 'history', 'examination', 'progress', 'discharge'
}
TOKENIZER = "cl100k_base"
MAX_MODEL_LEN = 2048
TARGET_CHUNK_TOKENS = 1000  # Reduced from 1200 to be more conservative
PROMPT_RESERVE = 400  # Increased buffer for prompt + response
MEDICAL_SECTION_HEADER = "=== MEDICAL SECTION ==="

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}")

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 count_tokens(text: str) -> int:
    encoding = tiktoken.get_encoding(TOKENIZER)
    return len(encoding.encode(text))

def extract_all_pages_with_token_count(file_path: str) -> Tuple[str, int, int]:
    try:
        text_chunks = []
        total_pages = 0
        total_tokens = 0
        
        with pdfplumber.open(file_path) as pdf:
            total_pages = len(pdf.pages)
            
            for i, page in enumerate(pdf.pages):
                page_text = page.extract_text() or ""
                lower_text = page_text.lower()
                
                if any(re.search(rf'\b{kw}\b', lower_text) for kw in MEDICAL_KEYWORDS):
                    section_header = f"\n{MEDICAL_SECTION_HEADER} (Page {i+1})\n"
                    text_chunks.append(section_header + page_text.strip())
                    total_tokens += count_tokens(section_header)
                else:
                    text_chunks.append(f"\n=== Page {i+1} ===\n{page_text.strip()}")
                
                total_tokens += count_tokens(page_text)
        
        return "\n".join(text_chunks), total_pages, total_tokens
    except Exception as e:
        return f"PDF processing error: {str(e)}", 0, 0

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, total_pages, total_tokens = extract_all_pages_with_token_count(file_path)
            result = json.dumps({
                "filename": os.path.basename(file_path),
                "content": text,
                "total_pages": total_pages,
                "total_tokens": total_tokens,
                "status": "complete"
            })
        elif file_type == "csv":
            chunks = []
            for chunk in pd.read_csv(file_path, encoding_errors="replace", header=None, dtype=str,
                                   skip_blank_lines=False, on_bad_lines="skip", chunksize=1000):
                chunks.append(chunk.fillna("").astype(str).values.tolist())
            content = [item for sublist in chunks for item in sublist]
            result = json.dumps({
                "filename": os.path.basename(file_path),
                "rows": content,
                "total_tokens": count_tokens(str(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,
                "total_tokens": count_tokens(str(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:
        return json.dumps({"error": f"Error processing {os.path.basename(file_path)}: {str(e)}"})

def clean_response(text: str) -> str:
    text = sanitize_utf8(text)
    text = re.sub(r"\[TOOL_CALLS\].*", "", text, flags=re.DOTALL)
    text = re.sub(r"\['get_[^\]]+\']\n?", "", text)
    text = re.sub(r"\{'meta':\s*\{.*?\}\s*,\s*'results':\s*\[.*?\]\}\n?", "", text, flags=re.DOTALL)
    text = re.sub(r"To analyze the medical records for clinical oversights.*?begin by reviewing.*?\n", "", text, flags=re.DOTALL)
    text = re.sub(r"\n{3,}", "\n\n", text).strip()
    return text

def format_final_report(analysis_results: List[str], filename: str) -> str:
    report = []
    report.append(f"COMPREHENSIVE CLINICAL OVERSIGHT ANALYSIS")
    report.append(f"Generated: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
    report.append(f"File: {filename}")
    report.append("=" * 80)
    
    sections = {
        "CRITICAL FINDINGS": [],
        "MISSED DIAGNOSES": [],
        "MEDICATION ISSUES": [],
        "ASSESSMENT GAPS": [],
        "FOLLOW-UP RECOMMENDATIONS": []
    }
    
    for result in analysis_results:
        for section in sections:
            section_match = re.search(
                rf"{re.escape(section)}:?\s*\n([^*]+?)(?=\n\*|\n\n|$)", 
                result, 
                re.IGNORECASE | re.DOTALL
            )
            if section_match:
                content = section_match.group(1).strip()
                if content and content not in sections[section]:
                    sections[section].append(content)
    
    if sections["CRITICAL FINDINGS"]:
        report.append("\n🚨 **CRITICAL FINDINGS** 🚨")
        for content in sections["CRITICAL FINDINGS"]:
            report.append(f"\n{content}")
    
    for section, contents in sections.items():
        if section != "CRITICAL FINDINGS" and contents:
            report.append(f"\n**{section.upper()}**")
            for content in contents:
                report.append(f"\n{content}")
    
    if not any(sections.values()):
        report.append("\nNo significant clinical oversights identified.")
    
    report.append("\n" + "=" * 80)
    report.append("END OF REPORT")
    
    return "\n".join(report)

def split_content_by_tokens(content: str, max_tokens: int = TARGET_CHUNK_TOKENS) -> List[str]:
    """More conservative splitting that ensures we stay well under token limits"""
    paragraphs = re.split(r"\n\s*\n", content)
    chunks = []
    current_chunk = []
    current_tokens = 0
    
    for para in paragraphs:
        para_tokens = count_tokens(para)
        
        # If paragraph is too big, split into sentences
        if para_tokens > max_tokens * 0.8:  # Don't allow paragraphs that take up most of the chunk
            sentences = re.split(r'(?<=[.!?])\s+', para)
            for sent in sentences:
                sent_tokens = count_tokens(sent)
                if current_tokens + sent_tokens > max_tokens * 0.9:  # Leave 10% buffer
                    if current_chunk:  # Only add if we have content
                        chunks.append("\n\n".join(current_chunk))
                        current_chunk = []
                        current_tokens = 0
                    
                    # If single sentence is too long, split into words
                    if sent_tokens > max_tokens * 0.8:
                        words = sent.split()
                        for word in words:
                            word_tokens = count_tokens(word)
                            if current_tokens + word_tokens > max_tokens * 0.9:
                                if current_chunk:
                                    chunks.append("\n\n".join(current_chunk))
                                    current_chunk = []
                                    current_tokens = 0
                            current_chunk.append(word)
                            current_tokens += word_tokens
                    else:
                        current_chunk.append(sent)
                        current_tokens += sent_tokens
                else:
                    current_chunk.append(sent)
                    current_tokens += sent_tokens
        elif current_tokens + para_tokens > max_tokens * 0.9:
            chunks.append("\n\n".join(current_chunk))
            current_chunk = [para]
            current_tokens = para_tokens
        else:
            current_chunk.append(para)
            current_tokens += para_tokens
    
    if current_chunk:
        chunks.append("\n\n".join(current_chunk))
    
    return chunks

def init_agent():
    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=2,
        seed=100,
        additional_default_tools=[],
    )
    agent.init_model()
    log_system_usage("After Load")
    print("✅ Agent Ready")
    return agent

def analyze_complete_document(content: str, filename: str, agent: TxAgent, temperature: float = 0.3) -> str:
    """Analyze complete document with strict token management"""
    chunks = split_content_by_tokens(content)
    analysis_results = []
    
    for i, chunk in enumerate(chunks):
        try:
            # Minimal prompt template
            base_prompt = """Analyze this medical content for:
1. Critical findings needing immediate attention
2. Potential missed diagnoses
3. Medication issues
4. Assessment gaps
5. Follow-up recommendations

Content:\n"""
            
            # Calculate available space
            prompt_tokens = count_tokens(base_prompt)
            max_content_tokens = MAX_MODEL_LEN - prompt_tokens - 300  # 300 tokens for response
            
            # Ensure chunk fits
            chunk_tokens = count_tokens(chunk)
            if chunk_tokens > max_content_tokens:
                # If still too big after splitting, truncate
                encoding = tiktoken.get_encoding(TOKENIZER)
                tokens = encoding.encode(chunk)
                chunk = encoding.decode(tokens[:max_content_tokens])
                print(f"Warning: Truncated chunk {i} from {chunk_tokens} to {max_content_tokens} tokens")
            
            prompt = base_prompt + chunk
            
            # Final verification
            total_tokens = count_tokens(prompt)
            if total_tokens > MAX_MODEL_LEN - 200:
                encoding = tiktoken.get_encoding(TOKENIZER)
                tokens = encoding.encode(prompt)
                prompt = encoding.decode(tokens[:MAX_MODEL_LEN - 200])
                print(f"Warning: Truncated final prompt from {total_tokens} tokens")
            
            response = ""
            for output in agent.run_gradio_chat(
                message=prompt,
                history=[],
                temperature=temperature,
                max_new_tokens=200,  # Conservative response length
                max_token=MAX_MODEL_LEN,
                call_agent=False,
                conversation=[],
            ):
                if output:
                    if isinstance(output, list):
                        for m in output:
                            if hasattr(m, 'content'):
                                response += clean_response(m.content)
                    elif isinstance(output, str):
                        response += clean_response(output)
            
            if response:
                analysis_results.append(response)
        except Exception as e:
            print(f"Error processing chunk {i}: {str(e)}")
            continue
    
    return format_final_report(analysis_results, filename)

def create_ui(agent):
    with gr.Blocks(
        theme=gr.themes.Soft(
            primary_hue="indigo",
            secondary_hue="blue",
            neutral_hue="slate",
            spacing_size="md",
            radius_size="md"
        ),
        title="Clinical Oversight Assistant",
        css="""
        .report-box {
            border: 1px solid #e0e0e0;
            border-radius: 8px;
            padding: 16px;
            background-color: #f9f9f9;
        }
        .file-upload {
            background-color: #f5f7fa;
            padding: 16px;
            border-radius: 8px;
        }
        .analysis-btn {
            width: 100%;
        }
        .critical-finding {
            color: #d32f2f;
            font-weight: bold;
        }
        .dataframe-container {
            height: 600px;
            overflow-y: auto;
        }
        """
    ) as demo:
        gr.Markdown("""
        <div style='text-align: center; margin-bottom: 20px;'>
            <h1 style='color: #2b3a67; margin-bottom: 8px;'>🩺 Clinical Oversight Assistant</h1>
            <p style='color: #5a6a8a; font-size: 16px;'>
                Analyze medical records for potential oversights and generate comprehensive reports
            </p>
        </div>
        """)

        with gr.Row(equal_height=False):
            with gr.Column(scale=1, min_width=400):
                with gr.Group(elem_classes="file-upload"):
                    file_upload = gr.File(
                        file_types=[".pdf", ".csv", ".xls", ".xlsx"],
                        file_count="multiple",
                        label="Upload Medical Records",
                        elem_id="file-upload"
                    )
                    with gr.Row():
                        clear_btn = gr.Button("Clear All", size="sm")
                        send_btn = gr.Button(
                            "Analyze Documents",
                            variant="primary",
                            elem_classes="analysis-btn"
                        )
                    
                    with gr.Accordion("Additional Options", open=False):
                        msg_input = gr.Textbox(
                            placeholder="Enter specific focus areas or questions...",
                            label="Analysis Focus",
                            lines=3
                        )
                        temperature = gr.Slider(
                            minimum=0.1,
                            maximum=1.0,
                            value=0.3,
                            step=0.1,
                            label="Analysis Strictness"
                        )
                
                status = gr.Textbox(
                    label="Processing Status",
                    interactive=False,
                    visible=True
                )

            with gr.Column(scale=2, min_width=600):
                with gr.Tabs():
                    with gr.TabItem("Analysis Report", id="report"):
                        report_output = gr.Textbox(
                            label="Clinical Oversight Findings",
                            lines=25,
                            max_lines=50,
                            interactive=False,
                            elem_classes="report-box"
                        )
                    
                    with gr.TabItem("Raw Data Preview", id="preview"):
                        with gr.Column(elem_classes="dataframe-container"):
                            data_preview = gr.Dataframe(
                                headers=["Page", "Content"],
                                datatype=["str", "str"],
                                interactive=False
                            )
                
                with gr.Row():
                    download_output = gr.File(
                        label="Download Full Report",
                        visible=True,
                        interactive=False
                    )
                    gr.Button("Save to EHR", visible=False)

        def analyze(files: List, message: str, temp: float):
            if not files:
                return (
                    {"value": "", "visible": True},
                    None,
                    {"value": "⚠️ Please upload at least one file to analyze.", "visible": True},
                    {"value": None, "visible": True}
                )
            
            yield (
                {"value": "", "visible": True},
                None,
                {"value": "⏳ Processing documents...", "visible": True},
                {"value": None, "visible": True}
            )
            
            file_contents = []
            filenames = []
            preview_data = []
            
            with ThreadPoolExecutor(max_workers=4) as executor:
                futures = []
                for f in files:
                    file_path = f.name
                    futures.append(executor.submit(
                        convert_file_to_json, 
                        file_path,
                        os.path.splitext(file_path)[1][1:].lower()
                    ))
                    filenames.append(os.path.basename(file_path))
                
                results = []
                for future in as_completed(futures):
                    result = sanitize_utf8(future.result())
                    try:
                        data = json.loads(result)
                        results.append(data)
                        if "content" in data:
                            preview_data.append([data["filename"], data["content"][:500] + "..."])
                    except Exception as e:
                        print(f"Error processing result: {e}")
                        continue
            
            yield (
                {"value": "", "visible": True},
                None,
                {"value": f"🔍 Analyzing {len(files)} documents...", "visible": True},
                {"value": preview_data[:20], "visible": True}
            )
            
            try:
                combined_content = "\n".join([
                    item.get("content", "") if isinstance(item, dict) and "content" in item 
                    else str(item.get("rows", "")) if isinstance(item, dict)
                    else str(item)
                    for item in results
                ])
                
                full_report = analyze_complete_document(
                    combined_content, 
                    " + ".join(filenames), 
                    agent,
                    temperature=temp
                )
                
                file_hash_value = hashlib.md5(combined_content.encode()).hexdigest()
                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_report)
                
                yield (
                    {"value": full_report, "visible": True},
                    report_path if os.path.exists(report_path) else None,
                    {"value": "✅ Analysis complete!", "visible": True},
                    {"value": preview_data[:20], "visible": True}
                )
            
            except Exception as e:
                error_msg = f"❌ Error during analysis: {str(e)}"
                print(error_msg)
                yield (
                    {"value": "", "visible": True},
                    None,
                    {"value": error_msg, "visible": True},
                    {"value": None, "visible": True}
                )

        send_btn.click(
            fn=analyze,
            inputs=[file_upload, msg_input, temperature],
            outputs=[report_output, download_output, status, data_preview],
            api_name="analyze"
        )
        
        clear_btn.click(
            fn=lambda: (
                None,
                None,
                "",
                None,
                {"value": 0.3},
                {"value": ""}
            ),
            inputs=None,
            outputs=[file_upload, download_output, status, data_preview, temperature, msg_input]
        )

    return demo

if __name__ == "__main__":
    print("🚀 Launching app...")
    try:
        import tiktoken
    except ImportError:
        print("Installing tiktoken...")
        subprocess.run([sys.executable, "-m", "pip", "install", "tiktoken"])
    
    agent = init_agent()
    demo = create_ui(agent)
    demo.queue(
        api_open=False,
        max_size=20
    ).launch(
        server_name="0.0.0.0",
        server_port=7860,
        show_error=True,
        allowed_paths=[report_dir],
        share=False
    )