Deploy Vivekananda scriptures semantic search Gradio app

app.py

@@ -0,0 +1,326 @@
+import os
+import json
+import gradio as gr
+import lancedb
+from huggingface_hub import snapshot_download
+from sentence_transformers import SentenceTransformer
+
+# 1. Setup paths and download dataset from Hugging Face Hub
+DB_REPO = "anurag-chand/vivekananda-scriptures-lancedb"
+LOCAL_DB_DIR = "./scriptures_lancedb"
+
+print(f"Checking for scriptures database locally at {LOCAL_DB_DIR}...")
+if not os.path.exists(LOCAL_DB_DIR) or not os.listdir(LOCAL_DB_DIR):
+    print(f"Database not found. Downloading {DB_REPO} from Hugging Face Hub...")
+    snapshot_download(
+        repo_id=DB_REPO,
+        repo_type="dataset",
+        local_dir=LOCAL_DB_DIR
+    )
+    print("Download complete!")
+
+# Connect to LanceDB
+db = lancedb.connect(LOCAL_DB_DIR)
+table = db.open_table("scriptures")
+print(f"Connected to scriptures table! Total records: {len(table):,}")
+
+# 2. Load model
+print("Loading Krutrim Vyakyarth model...")
+model = SentenceTransformer("krutrim-ai-labs/Vyakyarth", device="cpu")
+print("Model loaded successfully.")
+
+# Custom CSS for gorgeous aesthetics (dark mode, glassmorphism, responsive grid)
+custom_css = """
+body {
+    background-color: #0d0f12 !important;
+    color: #e2e8f0 !important;
+    font-family: 'Inter', sans-serif !important;
+}
+.gradio-container {
+    max-width: 1100px !important;
+    margin: 0 auto !important;
+    padding: 20px !important;
+    background-color: #0d0f12 !important;
+}
+.header-box {
+    text-align: center;
+    padding: 30px 20px;
+    background: linear-gradient(135deg, rgba(30, 41, 59, 0.5), rgba(15, 23, 42, 0.8));
+    border: 1px solid rgba(255, 255, 255, 0.05);
+    border-radius: 16px;
+    margin-bottom: 30px;
+    box-shadow: 0 4px 30px rgba(0, 0, 0, 0.3);
+}
+.header-box h1 {
+    font-size: 2.5em;
+    font-weight: 800;
+    background: linear-gradient(to right, #ffd700, #ff8c00);
+    -webkit-background-clip: text;
+    -webkit-text-fill-color: transparent;
+    margin-bottom: 10px;
+}
+.header-box p {
+    color: #94a3b8;
+    font-size: 1.1em;
+}
+.search-btn {
+    background: linear-gradient(135deg, #ff8c00, #d35400) !important;
+    color: white !important;
+    font-weight: bold !important;
+    border: none !important;
+    border-radius: 8px !important;
+    transition: all 0.3s ease !important;
+}
+.search-btn:hover {
+    transform: translateY(-1px) !important;
+    box-shadow: 0 4px 15px rgba(211, 84, 0, 0.4) !important;
+}
+.result-card {
+    background: rgba(30, 41, 59, 0.4);
+    border: 1px solid rgba(255, 255, 255, 0.06);
+    border-radius: 14px;
+    padding: 22px;
+    margin-bottom: 20px;
+    box-shadow: 0 4px 20px rgba(0, 0, 0, 0.15);
+    transition: all 0.3s ease;
+}
+.result-card:hover {
+    transform: translateY(-2px);
+    border-color: rgba(255, 140, 0, 0.3);
+    box-shadow: 0 8px 30px rgba(0, 0, 0, 0.3);
+}
+.card-header {
+    display: flex;
+    justify-content: space-between;
+    align-items: center;
+    border-bottom: 1px solid rgba(255, 255, 255, 0.06);
+    padding-bottom: 10px;
+    margin-bottom: 15px;
+    flex-wrap: wrap;
+    gap: 10px;
+}
+.match-badge {
+    background: linear-gradient(135deg, #ff8c00, #e67e22);
+    color: white;
+    padding: 4px 12px;
+    border-radius: 20px;
+    font-size: 0.85em;
+    font-weight: bold;
+}
+.meta-tags {
+    display: flex;
+    gap: 8px;
+    flex-wrap: wrap;
+}
+.meta-tag {
+    background: rgba(255, 255, 255, 0.05);
+    border: 1px solid rgba(255, 255, 255, 0.08);
+    color: #cbd5e1;
+    padding: 3px 10px;
+    border-radius: 6px;
+    font-size: 0.85em;
+}
+.shloka-section {
+    background: rgba(255, 215, 0, 0.03);
+    border-left: 4px solid #ffd700;
+    padding: 12px 16px;
+    margin-bottom: 15px;
+    border-radius: 0 8px 8px 0;
+}
+.shloka-text {
+    font-size: 1.25em;
+    color: #f1c40f;
+    line-height: 1.6;
+    margin: 0;
+    font-weight: bold;
+}
+.translation-section {
+    background: rgba(46, 204, 113, 0.03);
+    border-left: 4px solid #2ecc71;
+    padding: 12px 16px;
+    margin-bottom: 15px;
+    border-radius: 0 8px 8px 0;
+}
+.translation-text {
+    font-size: 1.05em;
+    color: #2ecc71;
+    line-height: 1.5;
+    margin: 0;
+    font-style: italic;
+}
+.commentary-section {
+    padding-left: 4px;
+}
+.commentary-header {
+    font-size: 0.95em;
+    font-weight: bold;
+    color: #94a3b8;
+    margin-bottom: 6px;
+    text-transform: uppercase;
+    letter-spacing: 0.05em;
+}
+.commentary-text {
+    font-size: 1.05em;
+    color: #cbd5e1;
+    line-height: 1.6;
+    margin: 0;
+}
+"""
+
+def semantic_search(query: str, limit: int = 5, min_similarity: float = 0.5) -> str:
+    if not query.strip():
+        return "<div style='text-align: center; color: #94a3b8; padding: 20px;'>Please enter a search query above.</div>"
+        
+    try:
+        # Encode query
+        query_vector = model.encode(query).tolist()
+        
+        # Search LanceDB table
+        results = table.search(query_vector).limit(limit).to_list()
+        
+        if not results:
+            return "<div style='text-align: center; color: #ff6b6b; padding: 20px;'>No results found. Try adjusting your query.</div>"
+            
+        html_output = ""
+        for idx, res in enumerate(results):
+            # Parse metadata
+            meta = {}
+            if "metadata" in res and res["metadata"]:
+                try:
+                    meta = json.loads(res["metadata"])
+                except Exception:
+                    pass
+                    
+            # Compute similarity from L2 distance
+            distance = res.get("_distance", 1.0)
+            similarity = max(0.0, 1.0 - (distance / 2.0))
+            
+            if similarity < min_similarity:
+                continue
+                
+            source = res.get("source_file") or meta.get("source_file") or "Unknown"
+            book = meta.get("book_title") or meta.get("title") or os.path.splitext(source)[0]
+            book = str(book).replace("_", " ").strip().title()
+            
+            chapter = meta.get("chapter_title") or meta.get("section") or meta.get("chapter_label") or ""
+            chapter = str(chapter).replace("_", " ").strip().title()
+            
+            shloka = meta.get("shloka") or meta.get("sutra") or ""
+            translation = meta.get("translation") or ""
+            commentary_author = meta.get("commentary_author") or ""
+            raw_text = res.get("text", "")
+            
+            # Format shloka block
+            shloka_html = ""
+            if shloka:
+                shloka_formatted = shloka.strip().replace("\n", "<br>")
+                shloka_html = f"""
+                <div class="shloka-section">
+                    <p class="shloka-text">{shloka_formatted}</p>
+                </div>
+                """
+                
+            # Format translation block
+            translation_html = ""
+            if translation:
+                translation_formatted = translation.strip().replace("\n", "<br>")
+                translation_html = f"""
+                <div class="translation-section">
+                    <p class="translation-text">{translation_formatted}</p>
+                </div>
+                """
+                
+            # Format commentary or text chunk
+            text_to_print = raw_text.strip()
+            if shloka and text_to_print.startswith(shloka):
+                # Clean out prepended shloka
+                parts = text_to_print.split("Commentary")
+                if len(parts) > 1:
+                    text_to_print = "Commentary" + " ".join(parts[1:])
+            
+            text_formatted = text_to_print.strip().replace("\n", "<br>")
+            
+            comm_title = f"Commentary ({commentary_author})" if commentary_author else "Scripture Text"
+            
+            html_output += f"""
+            <div class="result-card">
+                <div class="card-header">
+                    <span class="match-badge">Similarity: {similarity*100:.1f}%</span>
+                    <div class="meta-tags">
+                        <span class="meta-tag">📖 {book}</span>
+                        {f'<span class="meta-tag">🔖 {chapter}</span>' if chapter else ''}
+                        <span class="meta-tag">📄 {source}</span>
+                    </div>
+                </div>
+                {shloka_html}
+                {translation_html}
+                <div class="commentary-section">
+                    <div class="commentary-header">{comm_title}</div>
+                    <p class="commentary-text">{text_formatted}</p>
+                </div>
+            </div>
+            """
+            
+        if not html_output:
+            return f"<div style='text-align: center; color: #ff6b6b; padding: 20px;'>No results matched the similarity threshold of {min_similarity*100:.0f}%. Try lowering it.</div>"
+            
+        return html_output
+        
+    except Exception as e:
+        return f"<div style='text-align: center; color: #ff6b6b; padding: 20px;'>Error during search: {e}</div>"
+
+# 3. Create Gradio Interface Block
+with gr.Blocks(css=custom_css, title="Vivekananda Scriptures Semantic Search") as demo:
+    # Header block
+    gr.HTML("""
+    <div class="header-box">
+        <h1>Swami Vivekananda Scriptures Semantic Search</h1>
+        <p>Search over 686,000+ chunks of Sanskrit scriptures, translations, and commentaries (Patanjali Yoga Sutras, Gaudapada Karika, Vivekananda lectures, and more) using high-precision SOTA semantic vector matching.</p>
+    </div>
+    """)
+    
+    with gr.Row():
+        with gr.Column(scale=4):
+            query_input = gr.Textbox(
+                label="Search Query", 
+                placeholder="Type your search here (e.g., liberation from the cycle of birth and death, control of mind, nature of Brahman)...",
+                lines=1
+            )
+        with gr.Column(scale=1):
+            search_button = gr.Button("Search", elem_classes=["search-btn"])
+            
+    with gr.Row():
+        with gr.Column(scale=1):
+            limit_slider = gr.Slider(
+                label="Number of Results", 
+                minimum=1, 
+                maximum=20, 
+                value=5, 
+                step=1
+            )
+        with gr.Column(scale=1):
+            similarity_slider = gr.Slider(
+                label="Minimum Similarity Threshold", 
+                minimum=0.0, 
+                maximum=1.0, 
+                value=0.35, 
+                step=0.05
+            )
+            
+    # Outputs block
+    results_output = gr.HTML(label="Search Results")
+    
+    # Event binds
+    search_button.click(
+        fn=semantic_search,
+        inputs=[query_input, limit_slider, similarity_slider],
+        outputs=results_output
+    )
+    query_input.submit(
+        fn=semantic_search,
+        inputs=[query_input, limit_slider, similarity_slider],
+        outputs=results_output
+    )
+    
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,8 @@
+lancedb
+sentence-transformers
+huggingface_hub
+gradio
+pandas
+jinja2
+requests
+torch --index-url https://download.pytorch.org/whl/cpu