feat: implement Gradio-based Memory Keeper interface with multi-modal AI integration

app.py

@@ -0,0 +1,295 @@
+import gradio as gr
+import spaces
+import requests
+import uuid
+import base64
+import json
+from pathlib import Path
+
+# ── CONFIG ────────────────────────────────────────────────────────────────────
+BUILD_PERSONA_URL = "https://sheikhmdrakib-career--build-persona.modal.run"
+CHAT_URL          = "https://sheikhmdrakib-career--chat.modal.run"
+TRANSCRIBE_URL    = "https://sheikhmdrakib-career--transcribe.modal.run"
+VISION_URL        = "https://sheikhmdrakib-career--describe-photo.modal.run"
+OCR_URL           = "https://sheikhmdrakib-career--ocr-document.modal.run"
+TTS_URL           = "https://sheikhmdrakib-career--text-to-speech.modal.run"
+LIST_PERSONAS_URL = "https://sheikhmdrakib-career--list-personas.modal.run"
+# ─────────────────────────────────────────────────────────────────────────────
+
+
+def encode_file(path):
+    with open(path, "rb") as f:
+        return base64.b64encode(f.read()).decode()
+
+
+@spaces.GPU
+def build_persona(name, relationship, text_input, photo_captions, voice_file, photo_files, scanned_files):
+    if not name.strip():
+        return "❌ Please enter the person's name.", None, gr.update()
+
+    texts = [t.strip() for t in text_input.strip().split("---") if t.strip()] if text_input.strip() else []
+    captions = [c.strip() for c in photo_captions.strip().split("\n") if c.strip()] if photo_captions.strip() else []
+    voice_transcripts = []
+
+    if not texts and not captions and voice_file is None and not photo_files and not scanned_files:
+        return "❌ Please provide at least one input.", None, gr.update()
+
+    # 1. Transcribe voice note (Cohere ASR)
+    if voice_file is not None:
+        try:
+            r = requests.post(TRANSCRIBE_URL, json={
+                "audio_b64": encode_file(voice_file),
+                "filename": Path(voice_file).name,
+            }, timeout=180)
+            transcript = r.json().get("transcript", "")
+            if transcript:
+                voice_transcripts.append(transcript)
+        except Exception as e:
+            return f"❌ Voice transcription failed: {e}", None, gr.update()
+
+    # 2. Describe uploaded photos (MiniCPM-V)
+    if photo_files:
+        for photo in photo_files:
+            try:
+                r = requests.post(VISION_URL, json={"image_b64": encode_file(photo)}, timeout=180)
+                desc = r.json().get("description", "")
+                if desc:
+                    captions.append(desc)
+            except Exception as e:
+                captions.append(f"[Photo description failed: {e}]")
+
+    # 3. OCR scanned letters (Nemotron Parse)
+    if scanned_files:
+        for scan in scanned_files:
+            try:
+                r = requests.post(OCR_URL, json={"image_b64": encode_file(scan)}, timeout=180)
+                ocr_text = r.json().get("text", "")
+                if ocr_text:
+                    texts.append(ocr_text)
+            except Exception as e:
+                texts.append(f"[OCR failed: {e}]")
+
+    # 4. Build persona (Qwen 32B)
+    persona_id = str(uuid.uuid4())[:8]
+    try:
+        r = requests.post(BUILD_PERSONA_URL, json={
+            "persona_id": persona_id, "name": name.strip(),
+            "relationship": relationship.strip(),
+            "texts": texts, "photo_captions": captions,
+            "voice_transcripts": voice_transcripts,
+        }, timeout=1200)
+        result = r.json()
+        if result.get("success"):
+            persona = result["persona"]
+            summary = f"""✅ **{name}'s memory has been preserved.**
+
+**Persona ID:** `{persona_id}`
+**Personality:** {', '.join(persona.get('personality_traits', [])[:3])}
+**Language:** {persona.get('language', 'Auto')}
+**Memories captured:** {len(persona.get('key_memories', []))}
+**Voice style:** {persona.get('voice_description', 'N/A')}
+
+Go to the **💬 Talk** tab and enter the Persona ID."""
+            return summary, persona_id, gr.update(value=persona_id)
+        else:
+            return f"❌ Error: {result}", None, gr.update()
+    except Exception as e:
+        return f"❌ Failed: {e}", None, gr.update()
+
+
+@spaces.GPU
+def chat_with_persona(persona_id, message, history, language, enable_voice):
+    history = history or []
+
+    if not persona_id.strip():
+        history = history + [{"role": "assistant", "content": "⚠️ Please enter a Persona ID first."}]
+        return "", history, None
+
+    if not message.strip():
+        return "", history, None
+
+    try:
+        r = requests.post(CHAT_URL, json={
+            "persona_id": persona_id.strip(),
+            "history": [{"role": m["role"], "content": m["content"]} for m in history],
+            "message": message.strip(),
+            "language": language,
+        }, timeout=180)
+        result = r.json()
+        response_text = result.get("text", result.get("response", "..."))
+        voice_desc = result.get("voice_description", "warm elderly voice")
+    except Exception as e:
+        response_text = f"⚠️ Error: {e}"
+        voice_desc = "warm elderly voice"
+
+    history = history + [
+        {"role": "user", "content": message},
+        {"role": "assistant", "content": response_text},
+    ]
+
+    # Generate voice response (VoxCPM2)
+    audio_path = None
+    if enable_voice:
+        try:
+            r = requests.post(TTS_URL, json={
+                "text": response_text,
+                "voice_description": voice_desc,
+            }, timeout=180)
+            if r.status_code == 200:
+                import tempfile
+                with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
+                    f.write(r.content)
+                    audio_path = f.name
+        except Exception:
+            pass
+
+    return "", history, audio_path
+
+
+def load_personas():
+    for attempt in range(2):
+        try:
+            r = requests.get(LIST_PERSONAS_URL, timeout=90)
+            personas = r.json().get("personas", [])
+            if not personas:
+                return "No personas saved yet."
+            lines = [f"**{p['name']}** ({p['relationship']}) — ID: `{p['id']}`" for p in personas]
+            return "\n\n".join(lines)
+        except Exception as e:
+            if attempt == 0:
+                continue
+            return f"⚠️ Modal is waking up, please try again in 30 seconds."
+
+
+# ── UI ────────────────────────────────────────────────────────────────────────
+
+css = """
+@import url('https://fonts.googleapis.com/css2?family=Lora:ital,wght@0,400;0,600;1,400&family=Source+Sans+3:wght@300;400;600&display=swap');
+* { box-sizing: border-box; }
+body, .gradio-container { background: #0e0b08 !important; font-family: 'Source Sans 3', sans-serif !important; color: #e8dcc8 !important; }
+.gradio-container { max-width: 900px !important; margin: 0 auto !important; }
+h1, h2, h3 { font-family: 'Lora', serif !important; color: #d4a96a !important; }
+.header-title { text-align: center; font-family: 'Lora', serif; font-size: 2.4em; color: #d4a96a; margin: 24px 0 4px 0; }
+.header-sub { text-align: center; color: #8a7560; font-size: 1em; margin-bottom: 28px; font-style: italic; }
+.divider { border: none; border-top: 1px solid #2a2015; margin: 20px 0; }
+label { color: #8a7560 !important; font-size: 0.85em !important; letter-spacing: 0.08em !important; text-transform: uppercase !important; }
+textarea, input[type="text"] { background: #1a1510 !important; border: 1px solid #3a2e1e !important; color: #e8dcc8 !important; border-radius: 6px !important; }
+.model-badge { display: inline-block; background: #1f1710; border: 1px solid #3a2e1e; border-radius: 4px; padding: 2px 8px; font-size: 0.75em; color: #8a7560; margin: 2px; }
+"""
+
+with gr.Blocks(title="Memory Keeper") as demo:
+
+    gr.HTML("""
+        <div class="header-title">🕯️ Memory Keeper</div>
+        <div class="header-sub">Preserve the voice of someone you love. Talk to them again.</div>
+        <hr class="divider">
+        <div style="text-align:center; margin-bottom:16px;">
+            <span class="model-badge">🧠 MiniCPM4.1-8B (8B)</span>
+            <span class="model-badge">👁️ MiniCPM-V 4.6 (8B)</span>
+            <span class="model-badge">🎤 Cohere Transcribe (2B)</span>
+            <span class="model-badge">📄 Nemotron Parse v1.2 (&lt;1B)</span>
+            <span class="model-badge">🌍 Tiny Aya Fire (3.35B)</span>
+            <span class="model-badge">🌊 Tiny Aya Water (3.35B)</span>
+            <span class="model-badge">🔊 VoxCPM2 (~1B)</span>
+            <span class="model-badge">Total: ~26.7B params</span>
+        </div>
+    """)
+
+    with gr.Tabs():
+
+        # ── TAB 1: PRESERVE ──
+        with gr.Tab("📜 Preserve a Memory"):
+            gr.HTML("<p style='color:#8a7560; font-style:italic; margin-bottom:16px;'>Upload letters, photos, voice notes, or scanned documents. Each is processed by a specialized AI model.</p>")
+
+            with gr.Row():
+                name_input = gr.Textbox(label="Their Name", value="Dadu")
+                relationship_input = gr.Textbox(label="Your Relationship", value="Grandfather")
+
+            text_input = gr.Textbox(
+                label="📝 Letters / Diary Entries / Writings",
+                value="Dear family,\n\nWork hard in your studies. Knowledge is the one thing no one can take from you. When I was young, I walked four miles to school every day — no shoes, no complaints. We were poor but we were honest.\n\nAlways be kind to your mother. She sacrifices more than you will ever know.\n\nSave your money. A good name lasts longer than any wealth.\n\nI am proud of all of you. Your Dadu misses your faces.\n---\nBeta, I heard you got good marks. This makes my heart so happy. Keep going.",
+                lines=6,
+            )
+
+            with gr.Row():
+                photo_files = gr.File(
+                    label="🖼️ Photos (MiniCPM-V 4.6 will describe them)",
+                    file_count="multiple", file_types=["image"],
+                )
+                scanned_files = gr.File(
+                    label="📄 Scanned Letters/Docs (Nemotron Parse OCR)",
+                    file_count="multiple", file_types=["image"],
+                )
+
+            photo_captions = gr.Textbox(
+                label="🖼️ Manual Photo Captions (optional, one per line)",
+                value="Old black and white photo, Dadu sitting on a wooden chair, wearing white panjabi, smiling warmly\nEid gathering, Dadu surrounded by grandchildren, laughing out loud\nDadu in the garden early morning, watering plants carefully",
+                lines=3,
+            )
+
+            voice_input = gr.Audio(
+                label="🎤 Voice Note (Cohere Transcribe ASR)",
+                type="filepath", sources=["upload", "microphone"],
+            )
+
+            build_btn = gr.Button("✨ Preserve Their Memory", variant="primary")
+            build_output = gr.Markdown()
+            persona_id_state = gr.State()
+            persona_id_hidden = gr.Textbox(visible=False)
+
+            build_btn.click(
+                fn=build_persona,
+                inputs=[name_input, relationship_input, text_input, photo_captions,
+                        voice_input, photo_files, scanned_files],
+                outputs=[build_output, persona_id_state, persona_id_hidden],
+                show_progress="full",
+            )
+
+        # ── TAB 2: TALK ──
+        with gr.Tab("💬 Talk to Them"):
+            gr.HTML("<p style='color:#8a7560; font-style:italic; margin-bottom:16px;'>Enter the Persona ID and start a conversation. Enable voice to hear them speak.</p>")
+
+            with gr.Row():
+                persona_id_input = gr.Textbox(label="Persona ID", placeholder="e.g. a3f9c2b1")
+                language_select = gr.Dropdown(
+                    label="Language", choices=["auto", "English", "Bengali", "Hindi", "Chinese", "Japanese", "Korean", "Thai"], value="auto",
+                )
+                enable_voice = gr.Checkbox(label="🔊 Voice Response (VoxCPM2)", value=False)
+
+            chatbot = gr.Chatbot(label="", height=420, placeholder="*Their words will appear here...*")
+
+            with gr.Row():
+                msg_input = gr.Textbox(label="Your message", value="What is the most important lesson in life?", lines=2, scale=4)
+                send_btn = gr.Button("Send →", variant="primary", scale=1)
+
+            voice_output = gr.Audio(label="🔊 Voice Response", visible=True, autoplay=True)
+            clear_btn = gr.Button("Clear conversation", variant="secondary", size="sm")
+            chat_history = gr.State([])
+
+            send_btn.click(
+                fn=chat_with_persona,
+                inputs=[persona_id_input, msg_input, chat_history, language_select, enable_voice],
+                outputs=[msg_input, chatbot, voice_output],
+            )
+            msg_input.submit(
+                fn=chat_with_persona,
+                inputs=[persona_id_input, msg_input, chat_history, language_select, enable_voice],
+                outputs=[msg_input, chatbot, voice_output],
+            )
+            clear_btn.click(lambda: ([], []), outputs=[chat_history, chatbot])
+
+        # ── TAB 3: SAVED ──
+        with gr.Tab("📁 Saved Memories"):
+            refresh_btn = gr.Button("🔄 Load Saved Memories", variant="secondary")
+            personas_output = gr.Markdown()
+            refresh_btn.click(fn=load_personas, outputs=personas_output)
+
+    gr.HTML("""
+        <hr class="divider">
+        <p style='text-align:center; color:#3a2e1e; font-size:0.8em; font-style:italic;'>
+            Built for Build Small Hackathon · 6 AI Models · Hosted on Modal + Hugging Face
+        </p>
+    """)
+
+if __name__ == "__main__":
+    demo.launch(css=css)

requirements.txt

@@ -0,0 +1,55 @@
+annotated-doc==0.0.4
+annotated-types==0.7.0
+anyio==4.13.0
+audioop-lts==0.2.2
+brotli==1.2.0
+certifi==2026.5.20
+charset-normalizer==3.4.7
+click==8.4.1
+colorama==0.4.6
+fastapi==0.136.3
+filelock==3.29.1
+fsspec==2026.4.0
+gradio==6.16.0
+gradio_client==2.5.0
+groovy==0.1.2
+h11==0.16.0
+hf-gradio==0.4.1
+hf-xet==1.5.0
+httpcore==1.0.9
+httpx==0.28.1
+huggingface_hub==1.18.0
+idna==3.18
+Jinja2==3.1.6
+markdown-it-py==4.2.0
+MarkupSafe==3.0.3
+mdurl==0.1.2
+numpy==2.4.6
+orjson==3.11.9
+packaging==26.2
+pandas==3.0.3
+pillow==12.2.0
+pydantic==2.13.4
+pydantic_core==2.46.4
+pydub==0.25.1
+Pygments==2.20.0
+python-dateutil==2.9.0.post0
+python-multipart==0.0.32
+pytz==2026.2
+PyYAML==6.0.3
+requests==2.34.2
+rich==15.0.0
+safehttpx==0.1.7
+semantic-version==2.10.0
+shellingham==1.5.4
+six==1.17.0
+starlette==1.2.1
+tomlkit==0.14.0
+tqdm==4.68.1
+typer==0.25.1
+typing-inspection==0.4.2
+typing_extensions==4.15.0
+tzdata==2026.2
+urllib3==2.7.0
+uvicorn==0.49.0
+spaces