ui/ui_core.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

# ✅ Fix: Add src to Python path
sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "src")))

from txagent.txagent import TxAgent

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

def clean_final_response(text: str) -> str:
    cleaned = text.replace("[TOOL_CALLS]", "").strip()
    responses = cleaned.split("[Final Analysis]")

    if len(responses) <= 1:
        return f"<div style='padding:1em;border:1px solid #ccc;border-radius:12px;color:#fff;background:#353F54;'><p>{cleaned}</p></div>"

    panels = []
    for i, section in enumerate(responses[1:], 1):
        final = section.strip()
        panels.append(
            f"<div style='background:#2B2B2B;color:#E0E0E0;border-radius:12px;margin-bottom:1em;border:1px solid #888;'>"
            f"<div style='font-size:1.1em;font-weight:bold;padding:0.75em;background:#3A3A3A;color:#fff;border-radius:12px 12px 0 0;'>🧠 Final Analysis #{i}</div>"
            f"<div style='padding:1em;line-height:1.6;'>{final.replace(chr(10), '<br>')}</div>"
            f"</div>"
        )
    return "".join(panels)

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

def convert_file_to_json(file_path: str, file_type: str) -> str:
    try:
        cache_dir = os.path.join("cache")
        os.makedirs(cache_dir, exist_ok=True)
        h = file_hash(file_path)
        cache_path = os.path.join(cache_dir, f"{h}.json")

        if os.path.exists(cache_path):
            return open(cache_path, "r", encoding="utf-8").read()

        if file_type == "csv":
            df = pd.read_csv(file_path, encoding_errors="replace", header=None, dtype=str, skip_blank_lines=False, on_bad_lines="skip")
        elif file_type in ["xls", "xlsx"]:
            try:
                df = pd.read_excel(file_path, engine="openpyxl", header=None, dtype=str)
            except:
                df = pd.read_excel(file_path, engine="xlrd", header=None, dtype=str)
        elif file_type == "pdf":
            with pdfplumber.open(file_path) as pdf:
                text = "\n".join([page.extract_text() or "" for page in pdf.pages])
            result = json.dumps({"filename": os.path.basename(file_path), "content": text.strip()})
            open(cache_path, "w", encoding="utf-8").write(result)
            return result
        else:
            return json.dumps({"error": f"Unsupported file type: {file_type}"})

        if df is None or df.empty:
            return json.dumps({"warning": f"No data extracted from: {file_path}"})

        df = df.fillna("")
        content = df.astype(str).values.tolist()
        result = json.dumps({"filename": os.path.basename(file_path), "rows": content})
        open(cache_path, "w", encoding="utf-8").write(result)
        return result
    except Exception as e:
        return json.dumps({"error": f"Error reading {os.path.basename(file_path)}: {str(e)}"})

def chunk_text(text: str, max_tokens: int = 6000) -> List[str]:
    chunks = []
    words = text.split()
    chunk = []
    token_count = 0
    for word in words:
        token_count += len(word) // 4 + 1
        if token_count > max_tokens:
            chunks.append(" ".join(chunk))
            chunk = [word]
            token_count = len(word) // 4 + 1
        else:
            chunk.append(word)
    if chunk:
        chunks.append(" ".join(chunk))
    return chunks

def create_ui(agent: TxAgent):
    with gr.Blocks(theme=gr.themes.Soft()) as demo:
        gr.Markdown("<h1 style='text-align: center;'>📋 CPS: Clinical Patient Support System</h1>")

        chatbot = gr.Chatbot(label="CPS Assistant", height=600, type="messages")
        file_upload = gr.File(
            label="Upload Medical File",
            file_types=[".pdf", ".txt", ".docx", ".jpg", ".png", ".csv", ".xls", ".xlsx"],
            file_count="multiple"
        )
        message_input = gr.Textbox(placeholder="Ask a biomedical question or just upload the files...", show_label=False)
        send_button = gr.Button("Send", variant="primary")
        conversation_state = gr.State([])

        def handle_chat(message: str, history: list, conversation: list, uploaded_files: list, progress=gr.Progress()):
            context = (
                "You are an expert clinical AI assistant reviewing medical form or interview data. "
                "Your job is to analyze this data and reason about any information or red flags that a human doctor might have overlooked. "
                "Provide a **detailed and structured response**, including examples, supporting evidence from the form, and clinical rationale for why these items matter. "
                "Ensure the output is informative and helpful for improving patient care. "
                "Do not hallucinate. Base the response only on the provided form content. "
                "End with a section labeled '[Final Analysis]' where you summarize key findings the doctor may have missed."
            )

            try:
                history.append({"role": "user", "content": message})
                history.append({"role": "assistant", "content": "⏳ Processing your request..."})
                yield history

                extracted_text = ""
                if uploaded_files and isinstance(uploaded_files, list):
                    for file in uploaded_files:
                        if not hasattr(file, 'name'):
                            continue
                        path = file.name
                        ext = path.split(".")[-1].lower()
                        json_text = convert_file_to_json(path, ext)
                        extracted_text += sanitize_utf8(json_text) + "\n"

                chunks = chunk_text(extracted_text.strip())

                def process_chunk(i, chunk):
                    chunked_prompt = (
                        f"{context}\n\n--- Uploaded File Content (Chunk {i+1}/{len(chunks)}) ---\n\n{chunk}\n\n"
                        f"--- End of Chunk ---\n\nNow begin your analysis:"
                    )
                    try:
                        generator = agent.run_gradio_chat(
                            message=chunked_prompt,
                            history=[],
                            temperature=0.3,
                            max_new_tokens=1024,
                            max_token=8192,
                            call_agent=False,
                            conversation=conversation,
                            uploaded_files=uploaded_files,
                            max_round=30
                        )
                        result = ""
                        for update in generator:
                            if update is None:
                                print(f"[Warning] Empty response in chunk {i+1}")
                                continue
                            if isinstance(update, str):
                                result += update
                            elif isinstance(update, list):
                                for msg in update:
                                    if hasattr(msg, 'content'):
                                        result += msg.content
                        return result if result.strip() else f"[Chunk {i+1}] ⚠️ No response received."
                    except Exception as err:
                        print(f"[Error in chunk {i+1}] {err}")
                        return f"[Chunk {i+1}] ❌ Failed to process due to error."

                with ThreadPoolExecutor(max_workers=min(8, len(chunks))) as executor:
                    futures = [executor.submit(process_chunk, i, chunk) for i, chunk in enumerate(chunks)]
                    results = [f.result() for f in as_completed(futures)]

                full_response = "\n\n".join(results)
                full_response = clean_final_response(full_response.strip())
                history[-1] = {"role": "assistant", "content": full_response}
                yield history

            except Exception as chat_error:
                print(f"Chat handling error: {chat_error}")
                history[-1] = {"role": "assistant", "content": "❌ An error occurred while processing your request."}
                yield history

        inputs = [message_input, chatbot, conversation_state, file_upload]
        send_button.click(fn=handle_chat, inputs=inputs, outputs=chatbot)
        message_input.submit(fn=handle_chat, inputs=inputs, outputs=chatbot)

        gr.Examples([
            ["Upload your medical form and ask what the doctor might've missed."],
            ["This patient was treated with antibiotics for UTI. What else should we check?"],
            ["Is there anything abnormal in the attached blood work report?"]
        ], inputs=message_input)

    return demo