app.py

#!/usr/bin/env python
"""
GritAI — Audio → Notes  (Project #7 of "30 AI Projects in 15 Days")

Upload a recording or record from the mic -> a clean TRANSCRIPT with per-minute [mm:ss] timestamps
(quotable by the minute) -> a short SUMMARY + a list of ACTION ITEMS (task / owner / due).

FREE tier: faster-whisper ("small") on HF ZeroGPU for transcription + one Claude (haiku) call with
forced tool-use for schema-valid {summary, action_items}. PRO tier ("GritAI Studio"): long meetings,
SPEAKER attribution (who said what), a searchable/quotable archive, and auto-generated briefing docs --
the composition of several 30-in-15 rungs, assembled later. Shown here as a waitlist.

Two-stage pipeline: transcription runs inside @spaces.GPU (GPU held only for that); the summarization
Claude call runs after, on CPU, so the GPU is released before the network call.
"""
import os

import gradio as gr

ON_SPACE = bool(os.environ.get("SPACE_ID"))   # True when running on a Hugging Face Space

WHISPER_SIZE = "base"           # int8 on CPU — fast + reliable for short clips, no CUDA libs needed
CLAUDE_MODEL = os.environ.get("ANTHROPIC_MODEL", "claude-haiku-4-5")
MAX_TRANSCRIPT_CHARS = 24000    # keep the summarize call comfortably in-context

_MODEL = None


def _get_model():
    """Construct the faster-whisper model on CPU (int8), cached. Runs anywhere — no GPU / CUDA libs."""
    global _MODEL
    if _MODEL is None:
        from faster_whisper import WhisperModel
        _MODEL = WhisperModel(WHISPER_SIZE, device="cpu", compute_type="int8")
    return _MODEL


# Warm the model at startup on the Space so the first request isn't slow.
if ON_SPACE:
    try:
        _get_model()
        print("[startup] faster-whisper (base, cpu/int8) loaded", flush=True)
    except Exception as _e:
        print(f"[startup] whisper load failed (will retry on first call): {_e}", flush=True)


def _mmss(seconds):
    seconds = int(seconds or 0)
    return f"{seconds // 60:02d}:{seconds % 60:02d}"


def _transcribe(audio_path):
    try:
        model = _get_model()
    except ImportError:
        raise gr.Error(
            "🎙️ Local UI preview — transcription runs on the Space (faster-whisper isn't installed here). "
            "The controls and Studio waitlist are live."
        )
    segments, info = model.transcribe(audio_path, vad_filter=True, beam_size=5)
    rows = [(seg.start, seg.end, seg.text.strip()) for seg in segments]
    return rows, getattr(info, "language", "?")


NOTES_TOOL = {
    "name": "emit_notes",
    "description": "Return a concise summary and the action items from a meeting/voice-memo transcript.",
    "input_schema": {
        "type": "object",
        "properties": {
            "summary": {"type": "string", "description": "3-5 sentence plain-English summary."},
            "action_items": {
                "type": "array",
                "items": {
                    "type": "object",
                    "properties": {
                        "task": {"type": "string"},
                        "owner": {"type": "string", "description": "Person responsible, or 'unassigned'."},
                        "due": {"type": "string", "description": "Due date/'when', or 'not stated'."},
                    },
                    "required": ["task"],
                },
            },
        },
        "required": ["summary", "action_items"],
    },
}


def _summarize(transcript):
    key = os.environ.get("ANTHROPIC_API_KEY")
    if not key:
        return "_(Set ANTHROPIC_API_KEY on the Space to get the summary + action items.)_", ""
    try:
        import anthropic
        client = anthropic.Anthropic(api_key=key)
        msg = client.messages.create(
            model=CLAUDE_MODEL,
            max_tokens=1024,
            tools=[NOTES_TOOL],
            tool_choice={"type": "tool", "name": "emit_notes"},
            messages=[{"role": "user", "content":
                       "Summarize this transcript and extract the action items.\n\n" + transcript[:MAX_TRANSCRIPT_CHARS]}],
        )
        data = next((b.input for b in msg.content if getattr(b, "type", "") == "tool_use"), {})
    except Exception as e:
        return f"_(Summary unavailable: {type(e).__name__})_", ""

    summary = data.get("summary", "").strip()
    items = data.get("action_items", []) or []
    if items:
        lines = []
        for it in items:
            task = (it.get("task") or "").strip()
            owner = (it.get("owner") or "unassigned").strip()
            due = (it.get("due") or "not stated").strip()
            lines.append(f"- **{task}** — _{owner}_ · {due}")
        actions_md = "\n".join(lines)
    else:
        actions_md = "_No action items found._"
    return summary or "_(no summary)_", actions_md


def process(audio_path, progress=gr.Progress(track_tqdm=True)):
    if not audio_path:
        raise gr.Error("Upload a file or record from the mic first.")
    rows, lang = _transcribe(audio_path)
    if not rows:
        raise gr.Error("No speech detected in that audio.")
    transcript = "\n".join(f"[{_mmss(s)}] {text}" for s, e, text in rows if text)
    summary, actions_md = _summarize(transcript)
    return transcript, summary, actions_md


# ----------------------------------------------------------------------------- UI
THEME = gr.themes.Base(
    primary_hue=gr.themes.colors.orange, neutral_hue=gr.themes.colors.slate,
    font=[gr.themes.GoogleFont("IBM Plex Sans"), "system-ui", "sans-serif"],
    font_mono=[gr.themes.GoogleFont("IBM Plex Mono"), "monospace"],
).set(
    body_background_fill="#0a0b0e", body_text_color="#eef0f4", body_text_color_subdued="#9aa1b0",
    background_fill_primary="#111318", background_fill_secondary="#0e1015",
    block_background_fill="#111318", block_border_color="#20242e",
    block_label_background_fill="#0e1015", block_label_text_color="#9aa1b0", block_title_text_color="#eef0f4",
    border_color_primary="#20242e", border_color_accent="#c9451f",
    input_background_fill="#0e1015", input_border_color="#2b303c", input_border_color_focus="#c9451f",
    button_primary_background_fill="#ff5a2a", button_primary_background_fill_hover="#ff7d54",
    button_primary_text_color="#140a06", button_primary_border_color="#c9451f",
    button_secondary_background_fill="#15181f", button_secondary_text_color="#eef0f4",
    button_secondary_border_color="#2b303c",
    color_accent="#ff5a2a", color_accent_soft="#2b1810", slider_color="#ff5a2a",
    checkbox_background_color_selected="#ff5a2a", checkbox_border_color_focus="#c9451f",
)
CSS = """
@import url('https://fonts.googleapis.com/css2?family=Archivo:wght@700;800;900&family=IBM+Plex+Mono:wght@400;500;600&family=IBM+Plex+Sans:wght@400;500;600&display=swap');
.gradio-container{background:#0a0b0e !important}
.gradio-container::before{content:"";position:fixed;inset:0;z-index:0;pointer-events:none;
 background:repeating-linear-gradient(90deg,transparent 0 63px,rgba(255,255,255,.018) 63px 64px),
  repeating-linear-gradient(0deg,transparent 0 63px,rgba(255,255,255,.018) 63px 64px),
  radial-gradient(120% 70% at 50% -10%,#141821 0%,#0a0b0e 55%)}
.main,.contain{position:relative;z-index:1}
#ghead{display:flex;align-items:center;gap:14px;padding:6px 2px 4px}
#ghead .co{font-family:'Archivo',sans-serif;font-weight:900;font-size:19px;letter-spacing:.5px;color:#eef0f4;line-height:1}
#ghead .co b{color:#ff5a2a}
#ghead .sub{font-family:'IBM Plex Mono',monospace;font-size:10.5px;letter-spacing:.28em;text-transform:uppercase;color:#5f6675;margin-top:5px}
#ghead .tags{margin-left:auto;display:flex;gap:8px}
#ghead .tag{font-family:'IBM Plex Mono',monospace;font-size:10px;letter-spacing:.14em;text-transform:uppercase;color:#9aa1b0;border:1px solid #2b303c;border-radius:2px;padding:5px 9px}
#ghead .tag.hot{color:#ff5a2a;border-color:#c9451f}
#ghero h1{font-family:'Archivo',sans-serif;font-weight:800;font-size:clamp(26px,4.2vw,40px);line-height:1.05;letter-spacing:-.02em;margin:14px 0 4px;color:#eef0f4}
#ghero h1 .g{color:#ff5a2a}
#ghero p{color:#9aa1b0;margin:0;font-size:14.5px}
footer{visibility:hidden}
button.selected{color:#ff5a2a !important}
.gritai-foot{text-align:center;color:#5f6675;font-size:.8rem;margin-top:16px;font-family:'IBM Plex Mono',monospace;letter-spacing:.05em}
.gritai-foot a{color:#9aa1b0;text-decoration:none}
.pro-card{border:1px solid #2b303c;border-left:2px solid #ff5a2a;border-radius:4px;padding:16px 18px;background:#111318}
.pro-card h3{color:#eef0f4}
"""

with gr.Blocks(theme=THEME, css=CSS, title="GritAI · Audio → Notes") as demo:
    gr.HTML('''<div id="ghead">
      <div style="width:34px;height:34px;flex:none"><svg width="34" height="34" viewBox="0 0 38 38" fill="none"><rect x="1" y="1" width="36" height="36" rx="2" stroke="#2b303c"/><path d="M27 12.5A9 9 0 1 0 28 22H19" stroke="#ff5a2a" stroke-width="2.4" stroke-linecap="square"/><rect x="10.5" y="10.5" width="3" height="3" fill="#ff5a2a"/></svg></div>
      <div><div class="co">GRIT<b>AI</b></div><div class="sub">Audio &rarr; Notes</div></div>
      <div class="tags"><span class="tag hot">&#9656; Live</span><span class="tag">SDVOSB</span><span class="tag">Lawton OK</span></div>
    </div>
    <div id="ghero"><h1>Audio in. <span class="g">Notes</span> out.</h1>
    <p>Timestamped transcript + summary + action items &middot; quotable by the minute &middot; Project #7 of 30 in 15 days.</p></div>''')

    with gr.Tab("Transcribe & summarize (free)"):
        with gr.Row():
            with gr.Column(scale=2):
                audio = gr.Audio(sources=["upload", "microphone"], type="filepath", label="Audio")
                go = gr.Button("Transcribe ▶", variant="primary")
                gr.Markdown("<small>Best on short clips (a few minutes). Long meetings + speaker labels "
                            "are the Pro tier.</small>")
            with gr.Column(scale=3):
                summary = gr.Markdown(label="Summary")
                actions = gr.Markdown(label="Action items")
        transcript = gr.Textbox(label="Transcript ( [mm:ss] per line )", lines=16, show_copy_button=True)

        go.click(process, inputs=[audio], outputs=[transcript, summary, actions], api_name="transcribe")

    with gr.Tab("GritAI Studio (Pro)"):
        gr.Markdown(
            "<div class='pro-card'>\n\n"
            "### 🚀 Beyond the free demo\n"
            "The free tier transcribes short clips and pulls a summary + action items. **GritAI Studio** "
            "turns audio into an actual knowledge base:\n\n"
            "- **Long meetings** — hours, not minutes.\n"
            "- **Speaker attribution** — who said what, quotable by **minute _and_ speaker**.\n"
            "- **Searchable & quotable archive** — find any moment across every recording, with citations.\n"
            "- **Auto-briefings** — one-click briefing docs generated from the transcript.\n\n"
            "*Built on our own stack — the composition of several GritAI tools into one briefing room.*\n\n"
            "</div>"
        )
        with gr.Row():
            email = gr.Textbox(label="Email", placeholder="you@team.com", scale=3)
            use_case = gr.Textbox(label="What would you use it for?", placeholder="e.g. weekly stand-ups", scale=4)
        join = gr.Button("Join the Studio waitlist", variant="primary")
        waitlist_msg = gr.Markdown(visible=False)

        def _join(email_v, use_case_v):
            if not email_v or "@" not in email_v:
                return gr.update(value="⚠️ Please enter a valid email.", visible=True)
            return gr.update(value="✅ You're on the list — we'll reach out when GritAI Studio opens. Thanks!",
                             visible=True)

        join.click(_join, inputs=[email, use_case], outputs=waitlist_msg)

    gr.HTML("<div class='gritai-foot'>faster-whisper + Claude &middot; "
            "<a href='https://gritai.solutions'>GRITAI SOLUTIONS</a> &middot; 30-in-15 &#8470;7</div>")

if __name__ == "__main__":
    demo.queue(max_size=20).launch(server_name="0.0.0.0", server_port=int(os.environ.get("PORT", 7860)))