app.py

import os, io, base64, json, time
from typing import Optional, Tuple
import requests
from PIL import Image, ImageOps, ImageDraw
import streamlit as st
from streamlit_mic_recorder import mic_recorder, speech_to_text
from gtts import gTTS

# -----------------------
# Config
# -----------------------
BACKEND = os.getenv("BACKEND_URL", "http://localhost:8000").rstrip("/")
TIMEOUT = 180

st.set_page_config(page_title="Realtime BI Assistant (Frontend)", layout="centered")

# -----------------------
# Helpers
# -----------------------
HF_TOKEN = os.getenv("HF_TOKEN")

with st.sidebar:
    st.subheader("Auth / Tokens")
    _tok = st.text_input("HF Token (optional)", value=HF_TOKEN or "", type="password", help="Used by backend to call Hugging Face APIs.")
    if _tok.strip():
        HF_TOKEN = _tok.strip()

def _headers():
    h = {"Accept": "application/json"}
    if HF_TOKEN:
        h["Authorization"] = f"Bearer {HF_TOKEN}"
    return h

def ping_backend() -> Tuple[bool, Optional[dict]]:
    try:
        r = requests.get(f"{BACKEND}/", timeout=10, headers=_headers())
        return r.ok, (r.json() if r.ok else None)
    except Exception as e:
        return False, {"error": str(e)}

def post_json(path: str, payload: dict) -> requests.Response:
    return requests.post(f"{BACKEND}{path}", json=payload, timeout=TIMEOUT, headers=_headers())

def post_multipart(path: str, files: dict, params: dict) -> requests.Response:
    return requests.post(f"{BACKEND}{path}", files=files, params=params, timeout=TIMEOUT, headers=_headers())

def pil_from_upload(file) -> Optional[Image.Image]:
    try:
        return Image.open(file).convert("RGB")
    except Exception:
        return None

def compress_and_b64(img: Image.Image, max_side: int = 1280, quality: int = 85):
    img = ImageOps.exif_transpose(img)
    w0, h0 = img.size
    scale = max(w0, h0) / max_side if max(w0, h0) > max_side else 1.0
    img_proc = img.resize((int(w0/scale), int(h0/scale))) if scale > 1.0 else img

    buf = io.BytesIO()
    img_proc.save(buf, format="JPEG", quality=quality, optimize=True)
    b64 = base64.b64encode(buf.getvalue()).decode()
    return b64, img_proc, (w0, h0), img_proc.size

def draw_bbox(img: Image.Image, bbox: list[int], color=(0, 255, 0), width: int = 4) -> Image.Image:
    out = img.copy()
    draw = ImageDraw.Draw(out)
    x1, y1, x2, y2 = bbox
    draw.rectangle([x1, y1, x2, y2], outline=color, width=width)
    return out

def tts_gtts_bytes(text: str, lang: str = "en", tld: str = "com", slow: bool = False) -> bytes:
    buf = io.BytesIO()
    gTTS(text=text, lang=lang, tld=tld, slow=slow).write_to_fp(buf)
    return buf.getvalue()

# --- Chat state helpers ---
if "chat" not in st.session_state:
    st.session_state.chat = []   # list of {"role": "user"|"assistant", "text": str}

def add_chat(role: str, text: str):
    st.session_state.chat.append({"role": role, "text": text})

def render_chat_transcript():
    st.subheader("🗨️ Conversation")
    for m in st.session_state.chat[-100:]:  # show last 100 turns
        with st.chat_message("user" if m["role"]=="user" else "assistant"):
            st.markdown(m["text"])

# -----------------------
# Small UI renderers (so we can reorder cleanly)
# -----------------------
def render_examples_buttons(key_prefix: str = "main"):
    cols = st.columns(2)
    examples = [
        "What is total sales (revenue) of Ramesh?",
        "Revenue for BLR on 2025-09-06",
        "Monthly revenue for Electronics in BLR for 2025-09",
        "Top 5 SKUs by revenue in HYD on 2025-09-06 (include category)",
        "Ramesh's total sales in NCR on 2025-09-06",
    ]
    for i, ex in enumerate(examples):
        if cols[i % 2].button(ex, key=f"{key_prefix}_ex_{i}"):
            st.session_state["q_text"] = ex
            st.rerun()


def render_bi_question_section(section_heading=True, key_prefix: str = "main"):
    if section_heading:
        st.subheader("3) Ask a BI question")

    with st.expander("Examples", expanded=False):
        render_examples_buttons(key_prefix=key_prefix)

    # Use a unique key for the textarea.
    default_q = st.session_state.get("q_text", "What is total sales (revenue) of Ramesh?")
    q_text = st.text_area("Your question", value=default_q, height=100,
                          key=f"{key_prefix}_q_textarea")

    with st.expander("Optional: visual context (JSON)", expanded=False):
        vis_str = st.text_area("visual_ctx", value="{}", height=80,
                               key=f"{key_prefix}_vis_text")

    try:
        visual_ctx = json.loads(vis_str) if vis_str.strip() else {}
    except Exception:
        visual_ctx = {}
        st.warning("`visual_ctx` is not valid JSON; ignored.")

    if st.button("Ask", key=f"{key_prefix}_ask"):
        payload = {
            "user_id": st.session_state.user_name or None,
            "text": q_text.strip(),
            "visual_ctx": visual_ctx,
        }
        try:
            with st.spinner("Querying…"):
                r = post_json("/query", payload)

            if r.ok:
                resp = r.json()
                answer = resp.get("answer_text", "")
                st.success(answer)
                st.session_state["last_answer_text"] = answer

                sqls = [c[4:] for c in resp.get("citations", [])
                        if isinstance(c, str) and c.startswith("sql:")]
                if sqls:
                    with st.expander("SQL used", expanded=True):
                        st.code(sqls[0], language="sql")
                        for s in sqls[1:]:
                            st.code(s, language="sql")

                if resp.get("metrics"):
                    with st.expander("Metrics", expanded=False):
                        st.json(resp["metrics"])
                if resp.get("chart_refs"):
                    with st.expander("Charts", expanded=False):
                        st.json(resp["chart_refs"])
                if "uncertainty" in resp:
                    st.caption(f"Uncertainty: {resp['uncertainty']:.2f}")
            else:
                try:
                    err = r.json()
                except Exception:
                    err = {"detail": r.text}
                st.error(f"Backend error {r.status_code}: {err.get('detail')}")
                if "SQLGenTool disabled" in str(err.get("detail", "")):
                    st.info("Add your Hugging Face token in the sidebar (or set the HF_TOKEN env var).")
        except Exception as e:
            st.error(f"Request error: {e}")

def render_voice_to_text():
    st.caption("Voice → Text (browser STT)")
    c1, c2 = st.columns([2, 1])
    with c1:
        st.write("Click to speak; recognized text will fill the question box.")
    with c2:
        stt_lang = st.selectbox("STT language", ["en", "hi"], index=0, key="stt_lang_dd")
        if "prev_stt_lang" not in st.session_state:
            st.session_state["prev_stt_lang"] = stt_lang
        elif st.session_state["prev_stt_lang"] != stt_lang:
            st.session_state["prev_stt_lang"] = stt_lang
            st.rerun()

    stt_text = speech_to_text(
        language=st.session_state.get("stt_lang_dd", "en"),
        use_container_width=True,
        just_once=True,
        start_prompt="🎙️ Start recording",
        stop_prompt="⏹️ Stop recording",
        key="stt_main_btn",
    )

    if stt_text:
        st.session_state["q_text"] = stt_text
        st.success(f"Recognized: {stt_text}")
        st.rerun()

    st.markdown("---")
    st.caption("Optional: record & play raw audio (no transcription)")
    rec = mic_recorder(
        start_prompt="🎙️ Start",
        stop_prompt="⏹️ Stop",
        just_once=True,
        key="mic_raw_btn",
    )
    if rec and rec.get("bytes"):
        st.audio(rec["bytes"], format="audio/wav")

def render_tts_controls():
    st.markdown("---")
    st.caption("Text → Voice (gTTS)")
    tts_lang = st.selectbox("TTS language", ["en", "hi"], index=0, key="tts_lang_dd")
    tld_label = st.selectbox(
        "Accent / region (tld)",
        ["Default (.com)", "India (.co.in)", "US (.us)", "UK (.co.uk)"],
        index=1,
        key="tts_tld_dd"
    )
    tld_map = {
        "Default (.com)": "com",
        "India (.co.in)": "co.in",
        "US (.us)": "us",
        "UK (.co.uk)": "co.uk",
    }
    
    if st.button("🔊 Speak last answer", key="tts_speak_btn"):
        ans = st.session_state.get("last_answer_text", "")
        if not ans.strip():
            st.warning("Ask a question first to generate an answer.")
        else:
            try:
                mp3 = tts_gtts_bytes(ans, lang=tts_lang, tld=tld_map[tld_label], slow=False)
                st.audio(mp3, format="audio/mp3")
            except Exception as e:
                st.error(f"TTS error: {e}")

# -----------------------
# UI
# -----------------------
st.title("Realtime BI Assistant")
st.caption("Face upsert/identify + BI Q&A (text) via your FastAPI backend")

ok, info = ping_backend()
status_col, url_col = st.columns([1,3])
with status_col:
    st.metric("Backend", "Online ✅" if ok else "Offline ❌")
with url_col:
    st.code(BACKEND, language="text")

if not ok and info:
    st.warning(f"Backend unreachable: {info}")

# Persist chosen user
if "user_name" not in st.session_state:
    st.session_state.user_name = "mohit"

# -----------------------
# 1) Bulk enroll via ZIP (Images/<UserName>/*)
# -----------------------
with st.expander("1) Bulk enroll via ZIP (Images/<UserName>/*)", expanded=False):
    zip_up = st.file_uploader("Upload ZIP", type=["zip"], key="zip_enroll")
    if st.button("Enroll ZIP"):
        if not zip_up:
            st.error("Please upload a ZIP.")
        else:
            try:
                with st.spinner("Uploading & enrolling…"):
                    files = {"zipfile_upload": ("enroll.zip", zip_up.read(), "application/zip")}
                    r = post_multipart("/enroll_zip", files=files, params={})
                if r.ok:
                    st.success("Enrollment complete ✅")
                    st.json(r.json())
                else:
                    st.error(f"Enrollment failed: {r.status_code}")
                    st.text(r.text)
            except Exception as e:
                st.error(f"Request error: {e}")

# -----------------------
# 2) Identify from image
# -----------------------
with st.expander("2) Identify from image", expanded=False):
    col_u, col_c = st.columns(2)
    with col_u:
        test_upload = st.file_uploader("Upload test image", type=["jpg","jpeg","png"], key="test_upload")
    with col_c:
        test_cam = st.camera_input("Or capture from camera", key="test_cam")

    test_img = None
    if test_cam is not None:
        test_img = pil_from_upload(test_cam)
    elif test_upload is not None:
        test_img = pil_from_upload(test_upload)

    def encode_for_backend(img: Image.Image):
        b64_out = compress_and_b64(img)
        if isinstance(b64_out, (tuple, list)):
            b64_str = b64_out[0]
        else:
            b64_str = b64_out

        if isinstance(b64_str, (bytes, bytearray)):
            b64_str = b64_str.decode("utf-8")

        if isinstance(b64_str, str) and b64_str.startswith("data:"):
            b64_str = b64_str.split(",", 1)[1]

        raw = base64.b64decode(b64_str.encode("utf-8"))
        sent_img = Image.open(io.BytesIO(raw)).convert("RGB")
        return b64_str, sent_img

    def draw_many(img: Image.Image, dets: list[dict]) -> Image.Image:
        out = img.copy()
        draw = ImageDraw.Draw(out)
        for d in dets:
            x1, y1, x2, y2 = [int(v) for v in d.get("bbox", [0, 0, 0, 0])]
            name = d.get("decision", "Unknown")
            score = float(d.get("best_score", 0.0))
            label = f"{name} ({score:.3f})"
            draw.rectangle([x1, y1, x2, y2], outline=(0, 255, 0), width=3)
            try:
                tb = draw.textbbox((x1, y1), label)
                tw, th = tb[2] - tb[0], tb[3] - tb[1]
            except Exception:
                tw, th = max(60, len(label) * 7), 14
            by1 = max(0, y1 - th - 6)
            draw.rectangle([x1, by1, x1 + tw + 6, y1], fill=(0, 0, 0))
            draw.text((x1 + 3, by1 + 2), label, fill=(0, 255, 0))
        return out

    if st.button("Identify"):
        if test_img is None:
            st.warning("Please provide an image first.")
        else:
            try:
                b64, sent_img = encode_for_backend(test_img)
                with st.spinner("Identifying…"):
                    r = post_json("/identify_many", {"image_b64": b64, "top_k": 3})
                if not r.ok:
                    st.error(f"Identify failed: {r.status_code}")
                    st.text(r.text)
                else:
                    data = r.json()
                    dets = data.get("detections", [])
                    st.caption(f"Faces found: {len(dets)}")
                    st.image(draw_many(sent_img, dets), use_container_width=True)
                    if dets:
                        with st.expander("Details"):
                            st.json(dets)
            except Exception as e:
                st.error(f"Request error: {e}")

# -----------------------
# 2.5) Voice mode (frontend-only) with requested order
# -----------------------
st.subheader("🎙️ Voice mode (optional)")
with st.expander("Speak your question / hear the answer", expanded=True):
    # (1) Voice → Text first
    render_voice_to_text()

    # (2) Ask a BI question (same logic as main section)
    render_bi_question_section(section_heading=False, key_prefix="voice")

    # (3) Listen response (TTS button)
    render_tts_controls()

# # Shortcut: Talk → Ask → Speak (kept intact)
# if st.button("🎛️ Talk → Ask → Speak", key="talk_ask_speak"):
#     spoken = speech_to_text(language=st.session_state.get("stt_lang_dd", "en"),
#                             use_container_width=True, just_once=True, key="stt_inline_btn",
#                             start_prompt="🎙️ Start", stop_prompt="⏹️ Stop")
#     if spoken:
#         st.session_state["q_text"] = spoken
#         payload = {"user_id": st.session_state.user_name or None, "text": spoken, "visual_ctx": {}}
#         with st.spinner("Querying…"):
#             r = post_json("/query", payload)
#         if r.ok:
#             resp = r.json()
#             answer = resp.get("answer_text", "")
#             st.session_state["last_answer_text"] = answer
#             st.success(answer)
#             # Speak immediately with current TTS prefs
#             tld_map = {
#                 "Default (.com)": "com",
#                 "India (.co.in)": "co.in",
#                 "US (.us)": "us",
#                 "UK (.co.uk)": "co.uk",
#             }
#             tts_lang = st.session_state.get("tts_lang_dd", "en")
#             tld_label = st.session_state.get("tts_tld_dd", "India (.co.in)")
#             mp3 = tts_gtts_bytes(answer, lang=tts_lang, tld=tld_map.get(tld_label, "co.in"), slow=False)
#             st.audio(mp3, format="audio/mp3")
#         else:
#             st.error(f"Backend error {r.status_code}")

# -----------------------
# 2.6) Talk → Ask → Speak (voice chat with transcript)
# -----------------------
st.subheader("🗣️ Talk → Ask → Speak")
c_left, c_right = st.columns([2, 3])
with c_left:
    st.caption("Press to speak; we'll answer, speak back, and log the chat below.")
with c_right:
    # voice settings reuse your TTS controls' state if present; else defaults
    tts_lang = st.session_state.get("tts_lang_dd", "en")
    tld_map = {"Default (.com)": "com", "India (.co.in)": "co.in", "US (.us)": "us", "UK (.co.uk)": "co.uk"}
    tld_label = st.session_state.get("tts_tld_dd", "India (.co.in)")

# Mic widget (one utterance per click)
spoken = speech_to_text(
    language=st.session_state.get("stt_lang_dd", "en"),
    use_container_width=True,
    just_once=True,
    start_prompt="🎙️ Speak",
    stop_prompt="⏹️ Stop",
    key="stt_conv_btn",
)

if spoken:
    user_text = spoken.strip()
    if user_text:
        add_chat("user", user_text)
        payload = {"user_id": st.session_state.user_name or None, "text": user_text, "visual_ctx": {}}
        with st.spinner("Thinking…"):
            r = post_json("/query", payload)
        if r.ok:
            resp = r.json()
            answer = resp.get("answer_text", "").strip()
            add_chat("assistant", answer or "_(no rows)_")
            st.session_state["last_answer_text"] = answer
            # speak the answer
            try:
                mp3 = tts_gtts_bytes(answer or "I have no rows to report.",
                                     lang=tts_lang,
                                     tld=tld_map.get(tld_label, "co.in"),
                                     slow=False)
                st.audio(mp3, format="audio/mp3")
            except Exception as e:
                st.error(f"TTS error: {e}")
        else:
            try:
                err = r.json()
            except Exception:
                err = {"detail": r.text}
            add_chat("assistant", f"Backend error {r.status_code}: {err.get('detail')}")
            st.error(f"Backend error {r.status_code}: {err.get('detail')}")

# Show running transcript
render_chat_transcript()

# -----------------------
# 3) Ask a BI question (also kept as a main section for non-voice users)
# -----------------------
# render_bi_question_section(section_heading=True, key_prefix="main")