app.py

import sys
import os
import pandas as pd
import json
import gradio as gr
from typing import List, Tuple, Dict, Any, Generator, Union
import hashlib
import shutil
import re
from datetime import datetime
import time

# Configuration and 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")

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

os.environ["HF_HOME"] = model_cache_dir
os.environ["TRANSFORMERS_CACHE"] = model_cache_dir

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
MAX_TOKENS = 32768
MAX_NEW_TOKENS = 2048


def clean_response(text: str) -> str:
    try:
        text = text.encode('utf-8', 'surrogatepass').decode('utf-8')
    except UnicodeError:
        text = text.encode('utf-8', 'replace').decode('utf-8')
    text = re.sub(r"\[.*?\]|\bNone\b", "", text, flags=re.DOTALL)
    text = re.sub(r"\n{3,}", "\n\n", text)
    text = re.sub(r"[^\n#\-\*\w\s\.,:\(\)]+", "", text)
    return text.strip()


def estimate_tokens(text: str) -> int:
    return len(text) // 3.5


def extract_text_from_excel(file_path: str) -> str:
    all_text = []
    xls = pd.ExcelFile(file_path)
    for sheet_name in xls.sheet_names:
        df = xls.parse(sheet_name)
        df = df.astype(str).fillna("")
        rows = df.apply(lambda row: " | ".join(row), axis=1)
        sheet_text = [f"[{sheet_name}] {line}" for line in rows]
        all_text.extend(sheet_text)
    return "\n".join(all_text)


def split_text_into_chunks(text: str, max_tokens: int = MAX_TOKENS) -> List[str]:
    lines = text.split("\n")
    chunks = []
    current_chunk = []
    current_tokens = 0

    for line in lines:
        tokens = estimate_tokens(line)
        if current_tokens + tokens > max_tokens:
            chunks.append("\n".join(current_chunk))
            current_chunk = [line]
            current_tokens = tokens
        else:
            current_chunk.append(line)
            current_tokens += tokens

    if current_chunk:
        chunks.append("\n".join(current_chunk))
    return chunks


def build_prompt_from_text(chunk: str) -> str:
    return f"""
### Unstructured Clinical Records

You are reviewing unstructured, mixed-format clinical documentation from various forms, tables, and sheets.

**Objective:** Identify patterns, missed diagnoses, inconsistencies, and follow-up gaps.

Here is the extracted content chunk:

{chunk}

Please analyze the above and provide:
- Diagnostic Patterns
- Medication Issues
- Missed Opportunities
- Inconsistencies
- Follow-up Recommendations
"""


def init_agent():
    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=4,
        seed=100,
        additional_default_tools=[]
    )
    agent.init_model()
    return agent


def stream_final_report(agent, file) -> Generator[Union[Dict[str, str], Tuple[List[Dict[str, str]], str]], None, None]:
    extracted_text = extract_text_from_excel(file.name)
    chunks = split_text_into_chunks(extracted_text)
    chunk_responses = []

    for chunk in chunks:
        prompt = build_prompt_from_text(chunk)
        response = ""
        for result in agent.run_gradio_chat(
            message=prompt,
            history=[],
            temperature=0.2,
            max_new_tokens=MAX_NEW_TOKENS,
            max_token=MAX_TOKENS,
            call_agent=False,
            conversation=[],
        ):
            if isinstance(result, str):
                response += result
            elif hasattr(result, "content"):
                response += result.content
            elif isinstance(result, list):
                for r in result:
                    if hasattr(r, "content"):
                        response += r.content
        chunk_responses.append(clean_response(response))

    final_prompt = "\n\n".join(chunk_responses) + "\n\nSummarize the key findings above."
    yield {"role": "user", "content": f"[Excel Uploaded: {file.name}]"}
    stream_text = ""
    for result in agent.run_gradio_chat(
        message=final_prompt,
        history=[],
        temperature=0.2,
        max_new_tokens=MAX_NEW_TOKENS,
        max_token=MAX_TOKENS,
        call_agent=False,
        conversation=[],
    ):
        if isinstance(result, str):
            stream_text += result
        elif hasattr(result, "content"):
            stream_text += result.content
        elif isinstance(result, list):
            for r in result:
                if hasattr(r, "content"):
                    stream_text += r.content
        yield {"role": "assistant", "content": clean_response(stream_text)}

    final_report = f"# \U0001f9e0 Final Patient Report\n\n{clean_response(stream_text)}"
    report_path = os.path.join(report_dir, f"report_{datetime.now().strftime('%Y%m%d_%H%M%S')}.md")
    with open(report_path, 'w') as f:
        f.write(final_report)

    yield [{"role": "user", "content": f"[Excel Uploaded: {file.name}]"}, {"role": "assistant", "content": final_report}], report_path


def create_ui(agent):
    with gr.Blocks(title="Patient History Chat") as demo:
        chatbot = gr.Chatbot(label="Clinical Assistant", show_copy_button=True, type="messages")
        file_upload = gr.File(label="Upload Excel File", file_types=[".xlsx"])
        analyze_btn = gr.Button("🧠 Analyze Patient History")
        report_output = gr.File(label="Download Report")

        analyze_btn.stream(
            fn=stream_final_report,
            inputs=[file_upload],
            outputs=[chatbot, report_output]
        )

    return demo


if __name__ == "__main__":
    try:
        agent = init_agent()
        demo = create_ui(agent)
        demo.launch(
            server_name="0.0.0.0",
            server_port=7860,
            show_error=True,
            allowed_paths=["/data/hf_cache/reports"]
        )
    except Exception as e:
        print(f"Error: {str(e)}")
        sys.exit(1)