Create app.py

app.py

@@ -0,0 +1,391 @@
+import os
+import json
+import shutil
+import gradio as gr
+from datetime import datetime
+from typing import List, Dict, Any, Optional
+
+# --- Agent Imports & Safe Fallbacks ---
+try:
+    from alz_companion.agent import (
+        bootstrap_vectorstore, make_rag_chain, answer_query, synthesize_tts,
+        transcribe_audio, detect_tags_from_query, describe_image, build_or_load_vectorstore,
+        _default_embeddings
+    )
+    from alz_companion.prompts import BEHAVIOUR_TAGS, EMOTION_STYLES
+    from langchain.schema import Document
+    from langchain_community.vectorstores import FAISS
+    AGENT_OK = True
+except Exception as e:
+    AGENT_OK = False
+    # Define all fallback functions and classes
+    def bootstrap_vectorstore(sample_paths=None, index_path="data/"): return object()
+    def build_or_load_vectorstore(docs, index_path, is_personal=False): return object()
+    def make_rag_chain(vs_general, vs_personal, **kwargs): return lambda q, **k: {"answer": f"(Demo) You asked: {q}", "sources": []}
+    def answer_query(chain, q, **kwargs): return chain(q, **kwargs)
+    def synthesize_tts(text: str, lang: str = "en"): return None
+    def transcribe_audio(filepath: str, lang: str = "en"): return "This is a transcribed message."
+    def detect_tags_from_query(query: str, behavior_options: list, emotion_options: list): return {"detected_behavior": "None", "detected_emotion": "None"}
+    def describe_image(image_path: str): return "This is a description of an image."
+    class Document:
+        def __init__(self, page_content, metadata):
+            self.page_content = page_content
+            self.metadata = metadata
+    class FAISS:
+        def __init__(self):
+            self.docstore = type('obj', (object,), {'_dict': {}})()
+    BEHAVIOUR_TAGS = {"None": []}
+    EMOTION_STYLES = {"None": {}}
+    print(f"WARNING: Could not import from alz_companion ({e}). Running in UI-only demo mode.")
+
+# --- Centralized Configuration ---
+CONFIG = {
+    "themes": ["All", "The Father", "Still Alice", "Away from Her", "General Caregiving"],
+    "roles": ["patient", "caregiver"],
+    "behavior_tags": ["None"] + list(BEHAVIOUR_TAGS.keys()),
+    "emotion_tags": ["None"] + list(EMOTION_STYLES.keys()),
+    "languages": {"English": "en", "Chinese": "zh", "Malay": "ms", "French": "fr", "Spanish": "es"},
+    "tones": ["warm", "neutral", "formal", "playful"]
+}
+
+# --- File Management & Vector Store Logic ---
+INDEX_BASE = os.getenv('INDEX_BASE', 'data')
+UPLOADS_BASE = os.path.join(INDEX_BASE, "uploads")
+PERSONAL_INDEX_PATH = os.path.join(INDEX_BASE, "personal_faiss_index")
+os.makedirs(UPLOADS_BASE, exist_ok=True)
+THEME_PATHS = {t: os.path.join(INDEX_BASE, f"faiss_index_{t.replace(' ', '').lower()}") for t in CONFIG["themes"]}
+vectorstores = {}
+personal_vectorstore = None
+
+def canonical_theme(tk: str) -> str: return tk if tk in CONFIG["themes"] else "All"
+def theme_upload_dir(theme: str) -> str:
+    p = os.path.join(UPLOADS_BASE, f"theme_{canonical_theme(theme).replace(' ', '').lower()}")
+    os.makedirs(p, exist_ok=True)
+    return p
+def load_manifest(theme: str) -> Dict[str, Any]:
+    p = os.path.join(theme_upload_dir(theme), "manifest.json")
+    if os.path.exists(p):
+        try:
+            with open(p, "r", encoding="utf-8") as f: return json.load(f)
+        except Exception: pass
+    return {"files": {}}
+def save_manifest(theme: str, man: Dict[str, Any]):
+    with open(os.path.join(theme_upload_dir(theme), "manifest.json"), "w", encoding="utf-8") as f: json.dump(man, f, indent=2)
+def list_theme_files(theme: str) -> List[tuple[str, bool]]:
+    man = load_manifest(theme)
+    base = theme_upload_dir(theme)
+    found = [(n, bool(e)) for n, e in man.get("files", {}).items() if os.path.exists(os.path.join(base, n))]
+    existing = {n for n, e in found}
+    for name in sorted(os.listdir(base)):
+        if name not in existing and os.path.isfile(os.path.join(base, name)): found.append((name, False))
+    man["files"] = dict(found)
+    save_manifest(theme, man)
+    return found
+def copy_into_theme(theme: str, src_path: str) -> str:
+    fname = os.path.basename(src_path)
+    dest = os.path.join(theme_upload_dir(theme), fname)
+    shutil.copy2(src_path, dest)
+    return dest
+def seed_files_into_theme(theme: str):
+    SEED_FILES = [
+        ("sample_data/caregiving_tips.txt", True),
+        ("sample_data/the_father_segments_tagged_with_emotion_hybrid.jsonl", True),
+        ("sample_data/still_alice_segments_tagged_with_emotion_hybrid.jsonl", True),
+        ("sample_data/away_from_her_segments_tagged_with_emotion_hybrid.jsonl", True)
+    ]
+    man, changed = load_manifest(theme), False
+    for path, enable in SEED_FILES:
+        if not os.path.exists(path): continue
+        fname = os.path.basename(path)
+        if not os.path.exists(os.path.join(theme_upload_dir(theme), fname)):
+            copy_into_theme(theme, path)
+            man["files"][fname] = bool(enable)
+            changed = True
+    if changed: save_manifest(theme, man)
+
+def ensure_index(theme='All'):
+    theme = canonical_theme(theme)
+    if theme in vectorstores: return vectorstores[theme]
+    upload_dir = theme_upload_dir(theme)
+    enabled_files = [os.path.join(upload_dir, n) for n, enabled in list_theme_files(theme) if enabled]
+    index_path = THEME_PATHS.get(theme)
+    vectorstores[theme] = bootstrap_vectorstore(sample_paths=enabled_files, index_path=index_path)
+    return vectorstores[theme]
+
+# --- Gradio Callbacks ---
+def collect_settings(*args):
+    keys = ["role", "patient_name", "caregiver_name", "tone", "language", "tts_lang", "temperature", "behaviour_tag", "emotion_tag", "active_theme", "tts_on", "debug_mode"]
+    return dict(zip(keys, args))
+
+def add_personal_knowledge(text_input, file_input, image_input):
+    global personal_vectorstore
+    if not any([text_input, file_input, image_input]):
+        return "Please provide text, a file, or an image to add."
+    content_text, content_source = "", ""
+    if text_input and text_input.strip():
+        content_text, content_source = text_input.strip(), "Text Input"
+    elif file_input:
+        content_text, content_source = transcribe_audio(file_input.name), os.path.basename(file_input.name)
+    elif image_input:
+        content_text, content_source = describe_image(image_input.name), "Image Input"
+    if not content_text:
+        return "Could not extract any text content to add."
+    print("Auto-tagging personal memory...")
+    behavior_options = CONFIG.get("behavior_tags", [])
+    emotion_options = CONFIG.get("emotion_tags", [])
+    detected_tags = detect_tags_from_query(content_text, behavior_options=behavior_options, emotion_options=emotion_options)
+    detected_behavior = detected_tags.get("detected_behavior")
+    detected_emotion = detected_tags.get("detected_emotion")
+    print(f"  ...Detected Behavior: {detected_behavior}, Emotion: {detected_emotion}")
+    metadata = {"source": content_source}
+    if detected_behavior and detected_behavior != "None":
+        metadata["behaviors"] = [detected_behavior.lower()]
+    if detected_emotion and detected_emotion != "None":
+        metadata["emotion"] = detected_emotion.lower()
+    doc_to_add = Document(page_content=content_text, metadata=metadata)
+    if personal_vectorstore is None:
+        personal_vectorstore = build_or_load_vectorstore([doc_to_add], PERSONAL_INDEX_PATH, is_personal=True)
+    else:
+        personal_vectorstore.add_documents([doc_to_add])
+    personal_vectorstore.save_local(PERSONAL_INDEX_PATH)
+    return f"Successfully added memory with tags (Behavior: {detected_behavior}, Emotion: {detected_emotion})"
+
+def save_chat_to_memory(chat_history):
+    global personal_vectorstore
+    if not chat_history:
+        return "Nothing to save."
+    formatted_chat = []
+    for message in chat_history:
+        role = "User" if message["role"] == "user" else "Assistant"
+        content = message["content"].strip()
+        if content.startswith("*(Auto-detected context:"):
+            continue
+        formatted_chat.append(f"{role}: {content}")
+    conversation_text = "\n".join(formatted_chat)
+    if not conversation_text:
+        return "No conversation content to save."
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    doc_to_add = Document(page_content=conversation_text, metadata={"source": f"Conversation saved on {timestamp}"})
+    if personal_vectorstore is None:
+        personal_vectorstore = build_or_load_vectorstore([doc_to_add], PERSONAL_INDEX_PATH, is_personal=True)
+    else:
+        personal_vectorstore.add_documents([doc_to_add])
+    personal_vectorstore.save_local(PERSONAL_INDEX_PATH)
+    print(f"Saved conversation to long-term memory.")
+    return f"Conversation from {timestamp} saved successfully to long-term memory!"
+
+def list_personal_memories():
+    global personal_vectorstore
+    if personal_vectorstore is None or not hasattr(personal_vectorstore.docstore, '_dict') or not personal_vectorstore.docstore._dict:
+        return gr.update(value=[["No memories to display", ""]]), gr.update(choices=["No memories to select"], value=None)
+    docs = list(personal_vectorstore.docstore._dict.values())
+    dataframe_data = [[doc.metadata.get('source', 'Unknown'), doc.page_content] for doc in docs]
+    dropdown_choices = [doc.page_content for doc in docs]
+    return gr.update(value=dataframe_data), gr.update(choices=dropdown_choices)
+
+def delete_personal_memory(memory_to_delete):
+    global personal_vectorstore
+    if personal_vectorstore is None or not memory_to_delete:
+        return "Knowledge base is empty or no memory selected."
+    all_docs = list(personal_vectorstore.docstore._dict.values())
+    docs_to_keep = [doc for doc in all_docs if doc.page_content != memory_to_delete]
+    if len(all_docs) == len(docs_to_keep):
+        return "Error: Could not find the selected memory to delete."
+    print(f"Deleting memory. {len(docs_to_keep)} memories remaining.")
+    if not docs_to_keep:
+        if os.path.isdir(PERSONAL_INDEX_PATH):
+            shutil.rmtree(PERSONAL_INDEX_PATH)
+        personal_vectorstore = build_or_load_vectorstore([], PERSONAL_INDEX_PATH, is_personal=True)
+    else:
+        # Rebuild and save the index
+        new_vs = FAISS.from_documents(docs_to_keep, _default_embeddings())
+        new_vs.save_local(PERSONAL_INDEX_PATH)
+        personal_vectorstore = new_vs
+    return "Successfully deleted memory. The list will now refresh."
+
+def chat_fn(user_text, audio_file, settings, chat_history):
+    global personal_vectorstore
+    question = (user_text or "").strip()
+    if audio_file and not question:
+        try:
+            voice_lang_name = settings.get("tts_lang", "English")
+            voice_lang_code = CONFIG["languages"].get(voice_lang_name, "en")
+            question = transcribe_audio(audio_file, lang=voice_lang_code)
+        except Exception as e:
+            err_msg = f"Audio Error: {e}" if settings.get("debug_mode") else "Sorry, I couldn't understand the audio."
+            chat_history.append({"role": "assistant", "content": err_msg})
+            return "", None, chat_history
+    if not question:
+        return "", None, chat_history
+    chat_history.append({"role": "user", "content": question})
+    manual_behavior_tag = settings.get("behaviour_tag")
+    manual_emotion_tag = settings.get("emotion_tag")
+    if manual_behavior_tag not in [None, "None"] or manual_emotion_tag not in [None, "None"]:
+        scenario_tag, emotion_tag = manual_behavior_tag, manual_emotion_tag
+    else:
+        behavior_options = CONFIG.get("behavior_tags", [])
+        emotion_options = CONFIG.get("emotion_tags", [])
+        detected_tags = detect_tags_from_query(question, behavior_options=behavior_options, emotion_options=emotion_options)
+        scenario_tag, emotion_tag = detected_tags.get("detected_behavior"), detected_tags.get("detected_emotion")
+        if (scenario_tag and scenario_tag != "None") or (emotion_tag and emotion_tag != "None"):
+            detected_msg = f"*(Auto-detected context: Behavior=`{scenario_tag}`, Emotion=`{emotion_tag}`)*"
+            chat_history.append({"role": "assistant", "content": detected_msg})
+    active_theme = settings.get("active_theme", "All")
+    vs_general = ensure_index(active_theme)
+    if personal_vectorstore is None:
+        personal_vectorstore = build_or_load_vectorstore([], PERSONAL_INDEX_PATH, is_personal=True)
+    rag_chain_settings = {"role": settings.get("role"), "temperature": settings.get("temperature"), "language": settings.get("language"), "patient_name": settings.get("patient_name"), "caregiver_name": settings.get("caregiver_name"), "tone": settings.get("tone"),}
+    chain = make_rag_chain(vs_general, personal_vectorstore, **rag_chain_settings)
+    if scenario_tag == "None": scenario_tag = None
+    if emotion_tag == "None": emotion_tag = None
+    simple_history = chat_history[:-1]
+    response = answer_query(chain, question, chat_history=simple_history, scenario_tag=scenario_tag, emotion_tag=emotion_tag)
+    answer = response.get("answer", "[No answer found]")
+    chat_history.append({"role": "assistant", "content": answer})
+    audio_out = None
+    if settings.get("tts_on") and answer:
+        tts_lang_code = CONFIG["languages"].get(settings.get("tts_lang"), "en")
+        audio_out = synthesize_tts(answer, lang=tts_lang_code)
+    from gradio import update
+    return "", (update(value=audio_out, visible=bool(audio_out))), chat_history
+
+def upload_knowledge(files, current_theme):
+    if not files: return "No files were selected to upload."
+    added = 0
+    for f in files:
+        try:
+            copy_into_theme(current_theme, f.name); added += 1
+        except Exception as e: print(f"Error uploading file {f.name}: {e}")
+    if added > 0 and current_theme in vectorstores: del vectorstores[current_theme]
+    return f"Uploaded {added} file(s). Refreshing file list..."
+def save_file_selection(current_theme, enabled_files):
+    man = load_manifest(current_theme)
+    for fname in man['files']: man['files'][fname] = fname in enabled_files
+    save_manifest(current_theme, man)
+    if current_theme in vectorstores: del vectorstores[current_theme]
+    return f"Settings saved. Index for theme '{current_theme}' will rebuild on the next query."
+def refresh_file_list_ui(current_theme):
+    files = list_theme_files(current_theme)
+    enabled = [f for f, en in files if en]
+    msg = f"Found {len(files)} file(s). {len(enabled)} enabled."
+    return gr.update(choices=[f for f, _ in files], value=enabled), msg
+def auto_setup_on_load(current_theme):
+    theme_dir = theme_upload_dir(current_theme)
+    if not os.listdir(theme_dir):
+        print("First-time setup: Auto-seeding sample data...")
+        seed_files_into_theme(current_theme)
+    all_settings = collect_settings("patient", "", "", "warm", "English", "English", 0.7, "None", "None", "All", True, False)
+    files_ui, status_msg = refresh_file_list_ui(current_theme)
+    return all_settings, files_ui, status_msg
+
+# --- UI Definition ---
+CSS = ".gradio-container { font-size: 14px; } #chatbot { min-height: 250px; } #audio_out audio { max-height: 40px; } #audio_in audio { max-height: 40px; padding: 0; }"
+
+with gr.Blocks(theme=gr.themes.Soft(), css=CSS) as demo:
+    settings_state = gr.State({})
+
+    with gr.Tab("Chat"):
+        user_text = gr.Textbox(show_label=False, placeholder="Type your message here...")
+        audio_in = gr.Audio(sources=["microphone"], type="filepath", label="Voice Input", elem_id="audio_in")
+        with gr.Row():
+            submit_btn = gr.Button("Send", variant="primary")
+            save_btn = gr.Button("Save to Memory")
+            clear_btn = gr.Button("Clear")
+        chat_status = gr.Markdown()
+        audio_out = gr.Audio(label="Response Audio", autoplay=True, visible=True, elem_id="audio_out")
+        chatbot = gr.Chatbot(elem_id="chatbot", label="Conversation", type="messages")
+
+    with gr.Tab("Personalize"):
+        with gr.Accordion("Add to Personal Knowledge Base", open=True):
+            gr.Markdown("Add personal notes, memories, or descriptions of people and places. You can also upload audio/video notes or images.")
+            with gr.Row():
+                with gr.Column(scale=2):
+                    personal_text = gr.Textbox(lines=5, label="Text Input", placeholder="e.g., 'My father's name is John. He loves listening to Frank Sinatra music.'")
+                with gr.Column(scale=1):
+                    personal_file = gr.File(label="Upload Audio/Video File")
+                    personal_image = gr.Image(type="filepath", label="Upload Image")
+            with gr.Row():
+                personal_add_btn = gr.Button("Add Knowledge to Memory", variant="primary")
+            personal_status = gr.Markdown()
+        with gr.Accordion("Manage Personal Knowledge", open=False):
+            personal_memory_display = gr.DataFrame(headers=["Source", "Content"], label="Saved Personal Memories", interactive=False, row_count=(5, "dynamic"))
+            with gr.Row():
+                personal_refresh_btn = gr.Button("Refresh Memories")
+            with gr.Row():
+                personal_delete_selector = gr.Dropdown(label="Select a memory to delete", scale=3, interactive=True)
+                personal_delete_btn = gr.Button("Delete Selected Memory", variant="stop", scale=1)
+            personal_delete_status = gr.Markdown()
+
+    with gr.Tab("Settings"):
+        with gr.Group():
+            gr.Markdown("## Conversation & Persona Settings")
+            with gr.Row():
+                role = gr.Radio(CONFIG["roles"], value="caregiver", label="Your Role")
+                temperature = gr.Slider(0.0, 1.2, value=0.7, step=0.1, label="Creativity")
+                tone = gr.Dropdown(CONFIG["tones"], value="warm", label="Response Tone")
+            with gr.Row():
+                patient_name = gr.Textbox(label="Patient's Name", placeholder="e.g., 'Dad' or 'John'")
+                caregiver_name = gr.Textbox(label="Caregiver's Name", placeholder="e.g., 'me' or 'Jane'")
+                behaviour_tag = gr.Dropdown(CONFIG["behavior_tags"], value="None", label="Behaviour Filter (Manual Override)")
+                emotion_tag = gr.Dropdown(CONFIG["emotion_tags"], value="None", label="Emotion Filter (Manual Override)")
+        with gr.Accordion("Language, Voice & Debugging", open=False):
+            language = gr.Dropdown(list(CONFIG["languages"].keys()), value="English", label="Response Language")
+            tts_lang = gr.Dropdown(list(CONFIG["languages"].keys()), value="English", label="Voice Language")
+            tts_on = gr.Checkbox(True, label="Enable Voice Response (TTS)")
+            debug_mode = gr.Checkbox(False, label="Show Debug Info")
+        gr.Markdown("--- \n ## General Knowledge Base Management")
+        active_theme = gr.Radio(CONFIG["themes"], value="All", label="Active Knowledge Theme")
+        with gr.Row():
+            with gr.Column(scale=1):
+                files_in = gr.File(file_count="multiple", file_types=[".jsonl", ".txt"], label="Upload Knowledge Files")
+                upload_btn = gr.Button("Upload to Theme", variant="secondary")
+                seed_btn = gr.Button("Import Sample Data", variant="secondary")
+            with gr.Column(scale=2):
+                mgmt_status = gr.Markdown()
+                files_box = gr.CheckboxGroup(choices=[], label="Enable Files for the Selected Theme")
+                with gr.Row():
+                    save_files_btn = gr.Button("Save Selection", variant="primary")
+                    refresh_btn = gr.Button("Refresh List")
+
+    # --- Event Wiring ---
+    all_settings_components = [role, patient_name, caregiver_name, tone, language, tts_lang, temperature, behaviour_tag, emotion_tag, active_theme, tts_on, debug_mode]
+    for component in all_settings_components:
+        component.change(fn=collect_settings, inputs=all_settings_components, outputs=settings_state)
+    
+    submit_btn.click(fn=chat_fn, inputs=[user_text, audio_in, settings_state, chatbot], outputs=[user_text, audio_out, chatbot])
+    save_btn.click(fn=save_chat_to_memory, inputs=[chatbot], outputs=[chat_status])
+    clear_btn.click(lambda: (None, None, [], None, "", ""), outputs=[user_text, audio_out, chatbot, audio_in, user_text, chat_status])
+    
+    personal_add_btn.click(fn=add_personal_knowledge, inputs=[personal_text, personal_file, personal_image], outputs=[personal_status]).then(lambda: (None, None, None), outputs=[personal_text, personal_file, personal_image])
+    personal_refresh_btn.click(fn=list_personal_memories, inputs=None, outputs=[personal_memory_display, personal_delete_selector])
+    personal_delete_btn.click(fn=delete_personal_memory, inputs=[personal_delete_selector], outputs=[personal_delete_status]).then(fn=list_personal_memories, inputs=None, outputs=[personal_memory_display, personal_delete_selector])
+    
+    upload_btn.click(upload_knowledge, inputs=[files_in, active_theme], outputs=[mgmt_status]).then(refresh_file_list_ui, inputs=[active_theme], outputs=[files_box, mgmt_status])
+    save_files_btn.click(save_file_selection, inputs=[active_theme, files_box], outputs=[mgmt_status])
+    seed_btn.click(seed_files_into_theme, inputs=[active_theme]).then(refresh_file_list_ui, inputs=[active_theme], outputs=[files_box, mgmt_status])
+    refresh_btn.click(refresh_file_list_ui, inputs=[active_theme], outputs=[files_box, mgmt_status])
+    active_theme.change(refresh_file_list_ui, inputs=[active_theme], outputs=[files_box, mgmt_status])
+    demo.load(auto_setup_on_load, inputs=[active_theme], outputs=[settings_state, files_box, mgmt_status])
+
+# --- Startup Logic ---
+def pre_load_indexes():
+    global personal_vectorstore
+    print("Pre-loading all knowledge base indexes at startup...")
+    for theme in CONFIG["themes"]:
+        print(f"  - Loading general index for theme: '{theme}'")
+        try:
+            ensure_index(theme)
+            print(f"    ...'{theme}' theme loaded successfully.")
+        except Exception as e:
+            print(f"    ...Error loading theme '{theme}': {e}")
+    print("  - Loading personal knowledge index...")
+    try:
+        personal_vectorstore = build_or_load_vectorstore([], PERSONAL_INDEX_PATH, is_personal=True)
+        print("    ...Personal knowledge loaded successfully.")
+    except Exception as e:
+        print(f"    ...Error loading personal knowledge: {e}")
+    print("All indexes loaded. Application is ready.")
+
+if __name__ == "__main__":
+    pre_load_indexes()
+    demo.queue().launch(debug=True)