app.py

import os
from typing import List, Dict, Tuple, Union

import gradio as gr
from openai import OpenAI

# Optional parsers
import pandas as pd
from pypdf import PdfReader
from docx import Document as DocxDocument


# ===============================
# Client
# ===============================
def get_client(key: str) -> OpenAI:
    key = (key or "").strip()
    if not key:
        raise gr.Error("Please enter your OpenAI API key.")
    return OpenAI(api_key=key)


# ===============================
# Chat (messages format + streaming)
# ===============================
def stream_chat(api_key: str, user_input: str, history: List[Dict]):
    client = get_client(api_key)
    history = history or []
    user_input = (user_input or "").strip()
    if not user_input:
        return history, history, gr.update(value="")

    msgs = history + [{"role": "user", "content": user_input}]
    try:
        stream = client.chat.completions.create(
            model="gpt-5",
            messages=msgs,
            stream=True,
        )
        acc = ""
        for chunk in stream:
            delta = chunk.choices[0].delta.content or ""
            acc += delta
            yield msgs + [{"role": "assistant", "content": acc}], msgs, gr.update(value="")
        final_hist = msgs + [{"role": "assistant", "content": acc}]
        yield final_hist, final_hist, gr.update(value="")
    except Exception as e:
        err = f"[Error] {e}"
        final_hist = msgs + [{"role": "assistant", "content": err}]
        yield final_hist, final_hist, gr.update(value="")


# ===============================
# Pro Brief – File ingestion
# ===============================
TEXT_EXTS = {".txt", ".md", ".markdown"}
DOCX_EXTS = {".docx"}
PDF_EXTS  = {".pdf"}
CSV_EXTS  = {".csv"}

def _ext(path: str) -> str:
    return os.path.splitext(path.lower())[1]

def _coerce_paths(files: List[Union[str, dict, gr.File]]) -> List[str]:
    """
    Gradio may send:
      - list[str] of absolute filepaths (when type='filepath')
      - list[dict] with {'name': '/tmp/..'} in some versions
      - list[gr.File] objects with .name
    Normalize to list[str] filepaths.
    """
    paths = []
    for f in files or []:
        if isinstance(f, str):
            paths.append(f)
        elif isinstance(f, dict) and "name" in f:
            paths.append(f["name"])
        else:
            # gr.File or other object with .name
            name = getattr(f, "name", None)
            if name:
                paths.append(name)
    return paths

def read_text_file(fp: str) -> str:
    try:
        with open(fp, "r", encoding="utf-8") as f:
            return f.read()
    except UnicodeDecodeError:
        with open(fp, "r", encoding="latin-1") as f:
            return f.read()

def read_pdf(fp: str) -> str:
    text = []
    with open(fp, "rb") as f:
        reader = PdfReader(f)
        for page in reader.pages:
            txt = page.extract_text() or ""
            text.append(txt)
    return "\n".join(text).strip()

def read_docx(fp: str) -> str:
    doc = DocxDocument(fp)
    return "\n".join([p.text for p in doc.paragraphs]).strip()

def summarize_csv(fp: str) -> str:
    # Robust CSV read with separator fallbacks
    read_attempts = [
        dict(),
        dict(sep=";"),
        dict(sep="\t"),
    ]
    last_err = None
    df = None
    for kwargs in read_attempts:
        try:
            df = pd.read_csv(fp, **kwargs)
            break
        except Exception as e:
            last_err = e
    if df is None:
        raise gr.Error(f"Could not read CSV: {last_err}")

    shape_info = f"Rows: {df.shape[0]}, Columns: {df.shape[1]}"
    cols = ", ".join([f"{c} ({str(df[c].dtype)})" for c in df.columns])
    try:
        desc = df.describe(include="all").transpose().fillna("").to_string()
    except Exception:
        desc = "(describe() failed for this CSV)"
    try:
        head = df.head(10).to_string(index=False)
    except Exception:
        head = "(preview failed)"

    return (
        "CSV SUMMARY\n"
        f"{shape_info}\n\n"
        f"COLUMNS & TYPES:\n{cols}\n\n"
        f"DESCRIBE():\n{desc}\n\n"
        f"FIRST 10 ROWS:\n{head}\n"
    )

def load_files(files: List[Union[str, dict, gr.File]], progress: gr.Progress) -> Tuple[str, List[str]]:
    paths = _coerce_paths(files)
    if not paths:
        raise gr.Error("Please upload at least one file (PDF, DOCX, TXT, MD, or CSV).")

    texts = []
    names = []
    for i, path in enumerate(paths, start=1):
        names.append(os.path.basename(path))
        ext = _ext(path)
        progress((i-0.5)/max(len(paths), 1), desc=f"Parsing {os.path.basename(path)}")
        if ext in TEXT_EXTS:
            texts.append(read_text_file(path))
        elif ext in PDF_EXTS:
            texts.append(read_pdf(path))
        elif ext in DOCX_EXTS:
            texts.append(read_docx(path))
        elif ext in CSV_EXTS:
            texts.append(summarize_csv(path))
        else:
            raise gr.Error(f"Unsupported file type: {ext}")
        progress(i/max(len(paths), 1), desc=f"Parsed {os.path.basename(path)}")
    return "\n\n-----\n\n".join(texts), names


# ===============================
# Pro Brief – Chunking & synthesis
# ===============================
def chunk_text(s: str, max_chars: int = 12000) -> List[str]:
    s = (s or "").strip()
    if not s:
        return []
    if len(s) <= max_chars:
        return [s]
    chunks = []
    start = 0
    while start < len(s):
        end = min(start + max_chars, len(s))
        cut = s.rfind("\n\n", start, end)
        if cut == -1 or cut <= start + 2000:
            cut = end
        chunks.append(s[start:cut])
        start = cut
    return chunks

def llm_summarize_chunks(client: OpenAI, chunks: List[str], mode: str, custom_note: str, progress: gr.Progress) -> List[str]:
    summaries = []
    total = len(chunks)
    if total == 0:
        return summaries

    mode_prompt = {
        "Executive Brief": (
            "Create a crisp executive brief with sections: Context, Key Findings, Metrics, Implications, Decisions Needed."
        ),
        "Action Items": (
            "Extract actionable tasks with owners (if available), deadlines (if implied), dependencies, and priority."
        ),
        "Risks & Mitigations": (
            "Identify key risks, likelihood, impact, and concrete mitigations. Include watchpoints and triggers."
        ),
        "Meeting Minutes": (
            "Produce clean, structured minutes: Attendees (if inferable), Agenda, Discussion, Decisions, Action Items."
        ),
        "JSON Summary": (
            "Return a compact JSON with keys: context, findings[], metrics{}, actions[], risks[], decisions[]."
        ),
    }[mode]

    for i, ch in enumerate(chunks, start=1):
        progress(0.2 + 0.6*(i-1)/max(total,1), desc=f"Summarizing chunk {i}/{total}")
        sys = "You are a senior analyst. Write succinctly; use bullet points where appropriate."
        usr = f"{mode_prompt}\n\n{('Additional guidance: ' + custom_note) if custom_note else ''}\n\n---\nSOURCE CHUNK {i}/{total}:\n{ch}\n"
        resp = client.chat.completions.create(
            model="gpt-5",
            messages=[{"role": "system", "content": sys},
                      {"role": "user", "content": usr}],
        )
        summaries.append(resp.choices[0].message.content.strip())
        progress(0.2 + 0.6*(i)/max(total,1), desc=f"Summarized chunk {i}/{total}")
    return summaries

def llm_synthesize_final(client: OpenAI, mode: str, names: List[str], partials: List[str], custom_note: str, progress: gr.Progress) -> str:
    progress(0.85, desc="Synthesizing final deliverable")
    sys = "You are a chief of staff producing board-ready output. Tight, accurate, and well-structured."
    corpus = "\n\n---\n\n".join([f"[PART {i+1}]\n{p}" for i, p in enumerate(partials)])
    usr = (
        f"Files analyzed: {', '.join(names)}\n\n"
        f"Mode: {mode}\n"
        f"{('Additional guidance: ' + custom_note) if custom_note else ''}\n\n"
        "Synthesize the PARTS into a single cohesive deliverable. If JSON mode, return only JSON."
        "\n\n---\nCORPUS (SUMMARIES):\n" + corpus
    )
    resp = client.chat.completions.create(
        model="gpt-5",
        messages=[{"role": "system", "content": sys},
                  {"role": "user", "content": usr}],
    )
    progress(0.98, desc="Finalizing")
    return resp.choices[0].message.content.strip()

def pro_brief(api_key: str, files: List[Union[str, dict, gr.File]], mode: str, custom_note: str):
    progress = gr.Progress(track_tqdm=False)
    client = get_client(api_key)

    # Stage 1: Load files
    progress(0.02, desc="Loading files")
    out = "🔎 **Loading files...**\n"
    yield out

    raw_text, names = load_files(files, progress)
    out += f"✅ Parsed {len(names)} file(s): {', '.join(names)}\n"
    yield out

    # Stage 2: Chunk
    progress(0.18, desc="Chunking text")
    chunks = chunk_text(raw_text, max_chars=12000)
    out += f"🧱 Created {len(chunks)} chunk(s) for analysis\n"
    yield out

    # Stage 3: Summarize chunks
    partials = llm_summarize_chunks(client, chunks, mode, custom_note, progress)
    out += f"🧠 Summarized {len(partials)} chunk(s)\n"
    yield out

    # Stage 4: Synthesize final
    final = llm_synthesize_final(client, mode, names, partials, custom_note, progress)

    # Done
    progress(1.0, desc="Done")
    if mode == "JSON Summary":
        yield "```json\n" + final + "\n```"
    else:
        yield final


# ===============================
# UI
# ===============================
with gr.Blocks(title="ZEN GPT-5 • Production Tools") as demo:
    gr.Markdown("### 🔐 Enter your OpenAI API key (not stored)")
    api_key = gr.Textbox(placeholder="sk-...", type="password", label="OpenAI API Key")

    with gr.Tab("💬 Chat"):
        chatbox = gr.Chatbot(label="GPT-5 Chat", height=420, type="messages")
        history_state = gr.State([])
        user_in = gr.Textbox(placeholder="Say hi…", label="Message")
        send_btn = gr.Button("Send", variant="primary")
        clear_btn = gr.Button("Clear Chat")

        send_btn.click(stream_chat, [api_key, user_in, history_state], [chatbox, history_state, user_in], queue=True)
        user_in.submit(stream_chat, [api_key, user_in, history_state], [chatbox, history_state, user_in], queue=True)
        clear_btn.click(lambda: ([], []), None, [chatbox, history_state])

    with gr.Tab("📄 Pro Brief (Docs → Executive Output)"):
        gr.Markdown(
            "Upload PDFs, DOCX, TXT, MD, or CSV. Get an **Executive Brief**, **Action Items**, "
            "**Risks & Mitigations**, **Meeting Minutes**, or a **JSON Summary**."
        )
        files = gr.File(label="Upload files", file_count="multiple", type="filepath")
        mode = gr.Radio(
            ["Executive Brief", "Action Items", "Risks & Mitigations", "Meeting Minutes", "JSON Summary"],
            value="Executive Brief",
            label="Output Mode",
        )
        custom = gr.Textbox(label="Optional guidance (tone, audience, focus areas)", lines=3,
                            placeholder="e.g., Board-ready; focus on budget impact and timeline risk.")
        run = gr.Button("Generate Pro Brief", variant="primary")
        out = gr.Markdown(label="Output", show_copy_button=True)

        # Connect generator: yields interim status + final report
        run.click(pro_brief, [api_key, files, mode, custom], out, queue=True)

    # Subtle program stamp
    gr.HTML(
        "<div style='text-align:right; font-size:12px; opacity:0.55; margin-top:10px;'>"
        "Module 3 – ZEN SDK Production"
        "</div>"
    )

# Enable queuing (progress & concurrency-friendly)
demo.queue(max_size=64).launch()