feat: Add text embedding endpoint for semantic product search

Add text_search Gradio endpoint alongside existing image search.
Uses model.encode_text() from Jina CLIP v2 for 512-dim text embeddings.
Backward compatible: image endpoint keeps api_name='predict'.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

app.py

@@ -1,6 +1,7 @@
 """
 Visual Search API - HuggingFace Space
 Returns embedding vector for external Pinecone queries
+Supports both image and text inputs (Jina CLIP v2 multimodal)
 """
 
 import os
@@ -29,7 +30,7 @@ def load_model():
     return model
 
 
-def get_embedding(image: Image.Image) -> list:
+def get_image_embedding(image: Image.Image) -> list:
     """Generate 512-dim embedding for an image."""
     m = load_model()
 
@@ -44,22 +45,35 @@ def get_embedding(image: Image.Image) -> list:
         return emb.tolist()
 
 
-def search(image):
-    """Return embedding vector as JSON."""
+def get_text_embedding(text: str) -> list:
+    """Generate 512-dim embedding for a text query."""
+    m = load_model()
+
+    with torch.no_grad():
+        emb = m.encode_text([text])
+        if hasattr(emb, 'cpu'):
+            emb = emb.cpu().numpy()
+        emb = emb.flatten()
+        emb = emb / np.linalg.norm(emb)
+        if len(emb) > 512:
+            emb = emb[:512]
+        return emb.tolist()
+
+
+def image_search(image):
+    """Return image embedding vector as JSON."""
     if image is None:
         return json.dumps({"error": "No image provided"})
 
     try:
-        print("Generating embedding...")
-        embedding = get_embedding(image)
-        print(f"Embedding generated: {len(embedding)} dimensions")
+        print("Generating image embedding...")
+        embedding = get_image_embedding(image)
+        print(f"Image embedding generated: {len(embedding)} dimensions")
 
-        # Return embedding as JSON
-        result = {
+        return json.dumps({
             "embedding": embedding,
             "dimensions": len(embedding)
-        }
-        return json.dumps(result, indent=2)
+        }, indent=2)
 
     except Exception as e:
         import traceback
@@ -67,14 +81,60 @@ def search(image):
         return json.dumps({"error": str(e)})
 
 
-# Gradio interface - returns embedding as JSON
-demo = gr.Interface(
-    fn=search,
-    inputs=gr.Image(type="pil", label="Upload Image"),
-    outputs=gr.Textbox(label="Embedding Vector (JSON)", lines=15),
-    title="Visual Search - Embedding Generator",
-    description="Upload an image to get its 512-dimensional CLIP embedding as JSON."
-)
+def text_search(text):
+    """Return text embedding vector as JSON."""
+    if not text or not text.strip():
+        return json.dumps({"error": "No text provided"})
+
+    try:
+        text = text.strip()[:200]
+        print(f"Generating text embedding for: {text}")
+        embedding = get_text_embedding(text)
+        print(f"Text embedding generated: {len(embedding)} dimensions")
+
+        return json.dumps({
+            "embedding": embedding,
+            "dimensions": len(embedding)
+        }, indent=2)
+
+    except Exception as e:
+        import traceback
+        traceback.print_exc()
+        return json.dumps({"error": str(e)})
+
+
+# Gradio Blocks with explicit api_name for stable endpoints
+# Image: /call/predict (backward compatible with existing image-search.py)
+# Text:  /call/text_search (new endpoint for text-search.py)
+with gr.Blocks(title="Visual Search - Embedding Generator") as demo:
+    gr.Markdown("# Visual Search - Embedding Generator")
+    gr.Markdown("Upload an image or enter text to get a 512-dimensional CLIP embedding.")
+
+    with gr.Tab("Image Search"):
+        image_input = gr.Image(type="pil", label="Upload Image")
+        image_output = gr.Textbox(label="Embedding Vector (JSON)", lines=15)
+        image_btn = gr.Button("Generate Embedding")
+        image_btn.click(
+            image_search,
+            inputs=image_input,
+            outputs=image_output,
+            api_name="predict"
+        )
+
+    with gr.Tab("Text Search"):
+        text_input = gr.Textbox(
+            label="Search Query",
+            placeholder="e.g. boys underwear",
+            lines=1
+        )
+        text_output = gr.Textbox(label="Embedding Vector (JSON)", lines=15)
+        text_btn = gr.Button("Generate Embedding")
+        text_btn.click(
+            text_search,
+            inputs=text_input,
+            outputs=text_output,
+            api_name="text_search"
+        )
 
 if __name__ == "__main__":
     demo.queue().launch()