Added moderation and tutorial

src/components.py

@@ -488,9 +488,21 @@ def create_header():
         dbc.Col([
             html.Div([
                 html.H1("galaxy semantic search", className="galaxy-title text-center mb-1"),
-                html.P("powered by AION-Search", className="text-center mb-2",
-                      style={"color": "rgba(245, 245, 247, 0.5)", "font-weight": "300",
-                            "font-size": "0.8rem", "letter-spacing": "0.05em"}),
+                html.P([
+                    "powered by ",
+                    html.A(
+                        "AION-Search",
+                        href="https://aion-search.github.io/",
+                        target="_blank",
+                        rel="noopener noreferrer",
+                        style={
+                            "color": "rgba(245, 245, 247, 0.5)",
+                            "textDecoration": "underline"
+                        }
+                    )
+                ], className="text-center mb-2",
+                   style={"color": "rgba(245, 245, 247, 0.5)", "font-weight": "300",
+                         "font-size": "0.8rem", "letter-spacing": "0.05em"}),
                 html.Div(id="galaxy-count", className="galaxy-count text-center")
             ], className="text-center mb-3")
         ])
@@ -533,13 +545,29 @@ def create_search_container():
     return dbc.Row([
         dbc.Col([
             html.Div([
-                # Info button in top right
+                # Info button and tutorial link in top right
                 html.Div([
+                    html.A(
+                        "Tutorial",
+                        href="https://blog.nolank.ca/aion-search/#best-practices",
+                        target="_blank",
+                        rel="noopener noreferrer",
+                        style={
+                            "textDecoration": "underline",
+                            "marginRight": "0.1rem",
+                            "color": "rgba(245, 245, 247, 0.6)",
+                            "fontSize": "1.0rem",
+                            "fontWeight": "600",
+                            "letterSpacing": "0.03em",
+                            "transition": "all 0.3s ease"
+                        }
+                    ),
                     dbc.Button([
                         html.I(className="fas fa-info-circle")
                     ], id="info-button", color="link", size="sm",
                     className="info-button")
-                ], style={"position": "absolute", "top": "8px", "right": "8px", "z-index": "1000"}),
+                ], style={"position": "absolute", "top": "8px", "right": "8px", "z-index": "1000",
+                          "display": "flex", "alignItems": "center"}),
 
                 # Example search buttons
                 html.Div([
@@ -886,9 +914,23 @@ def create_info_modal():
             html.P("Images are from DESI Legacy Surveys DR10 via the hips2fits service provided by the Strasbourg Astronomical Data Centre (CDS).",
                    style={"color": "rgba(245, 245, 247, 0.6)", "margin-bottom": "0", "font-size": "0.75rem"})
         ]),
-        dbc.ModalFooter(
+        dbc.ModalFooter([
+            html.A(
+                dbc.Button([html.I(className="fas fa-file-alt me-2"), "Paper"], color="secondary"),
+                href="https://arxiv.org/abs/2512.11982",
+                target="_blank",
+                rel="noopener noreferrer",
+                className="me-2"
+            ),
+            html.A(
+                dbc.Button([html.I(className="fas fa-book me-2"), "Tutorial"], color="secondary"),
+                href="https://blog.nolank.ca/aion-search/#best-practices",
+                target="_blank",
+                rel="noopener noreferrer",
+                className="me-2"
+            ),
             dbc.Button("Close", id="close-info-modal", className="ms-auto")
-        )
+        ])
     ], id="info-modal", size="lg", is_open=False)
 
 

src/services.py

@@ -185,6 +185,29 @@ class EmbeddingService:
             self.openai_client = OpenAI(api_key=OPENAI_API_KEY)
         return self.openai_client
 
+    def _moderate_content(self, text: str) -> bool:
+        """Check if text content is appropriate using OpenAI Moderation API.
+
+        Args:
+            text: Text to moderate
+
+        Returns:
+            True if content is safe, False if flagged
+        """
+        try:
+            client = self._get_openai_client()
+            response = client.moderations.create(input=text)
+
+            # If any category is flagged, reject the content
+            if response.results[0].flagged:
+                logger.warning(f"Content moderation flagged input")
+                return False
+            return True
+        except Exception as e:
+            logger.error(f"Moderation API error: {e}")
+            # On error, allow the content through (fail open)
+            return True
+
     def encode_text_query(self, query: str) -> np.ndarray:
         """Encode text query using OpenAI embeddings + CLIP text projector.
 
@@ -194,6 +217,10 @@ class EmbeddingService:
         Returns:
             CLIP embedding vector
         """
+        # Moderate content first
+        if not self._moderate_content(query):
+            raise ValueError("Content moderation filter triggered")
+
         client = self._get_openai_client()
 
         # Get OpenAI text embedding
@@ -220,6 +247,11 @@ class EmbeddingService:
         Returns:
             Combined normalized embedding vector
         """
+        # Moderate all queries first
+        for query in queries:
+            if not self._moderate_content(query):
+                raise ValueError("Content moderation filter triggered")
+
         client = self._get_openai_client()
 
         # Get all embeddings at once for efficiency
@@ -426,14 +458,21 @@ class SearchService:
         Returns:
             DataFrame with search results
         """
-        # Encode query
-        query_embedding = self.embedding_service.encode_text_query(query)
-
-        # Build filter
-        filter_expr = self._build_rmag_filter(rmag_min, rmag_max)
-
-        # Search Zilliz
-        return self.zilliz_service.search(query_embedding, top_k, filter_expr)
+        try:
+            # Encode query
+            query_embedding = self.embedding_service.encode_text_query(query)
+
+            # Build filter
+            filter_expr = self._build_rmag_filter(rmag_min, rmag_max)
+
+            # Search Zilliz
+            return self.zilliz_service.search(query_embedding, top_k, filter_expr)
+        except ValueError as e:
+            # Content moderation triggered - return empty results silently
+            if "moderation" in str(e).lower():
+                logger.info("Search blocked by content moderation")
+                return pd.DataFrame()
+            raise
 
     def search_vector(
         self,
@@ -455,14 +494,21 @@ class SearchService:
         Returns:
             DataFrame with search results
         """
-        # Encode and combine vectors
-        combined_embedding = self.embedding_service.encode_vector_queries(queries, operations)
-
-        # Build filter
-        filter_expr = self._build_rmag_filter(rmag_min, rmag_max)
-
-        # Search Zilliz
-        return self.zilliz_service.search(combined_embedding, top_k, filter_expr)
+        try:
+            # Encode and combine vectors
+            combined_embedding = self.embedding_service.encode_vector_queries(queries, operations)
+
+            # Build filter
+            filter_expr = self._build_rmag_filter(rmag_min, rmag_max)
+
+            # Search Zilliz
+            return self.zilliz_service.search(combined_embedding, top_k, filter_expr)
+        except ValueError as e:
+            # Content moderation triggered - return empty results silently
+            if "moderation" in str(e).lower():
+                logger.info("Search blocked by content moderation")
+                return pd.DataFrame()
+            raise
 
     def search_advanced(
         self,
@@ -488,51 +534,58 @@ class SearchService:
         Returns:
             DataFrame with search results
         """
-        combined_embedding = None
-
-        # Process text queries
-        if text_queries and len(text_queries) > 0:
-            for query, weight in zip(text_queries, text_weights):
-                query_embedding = self.embedding_service.encode_text_query(query)
-
-                # Apply weight
-                weighted_embedding = query_embedding * weight
-
-                if combined_embedding is None:
-                    combined_embedding = weighted_embedding
-                else:
-                    combined_embedding += weighted_embedding
-
-        # Process image queries
-        if image_queries and len(image_queries) > 0:
-            if self.image_service is None:
-                raise RuntimeError("Image service not initialized")
-
-            for img_query, weight in zip(image_queries, image_weights):
-                # Encode image
-                image_embedding = self.image_service.encode_image(
-                    ra=img_query['ra'],
-                    dec=img_query['dec'],
-                    fov=img_query.get('fov', 0.025),
-                    size=256
-                )
-
-                # Apply weight
-                weighted_embedding = image_embedding * weight
-
-                if combined_embedding is None:
-                    combined_embedding = weighted_embedding
-                else:
-                    combined_embedding += weighted_embedding
+        try:
+            combined_embedding = None
 
-        # Normalize the final combined embedding
-        if combined_embedding is not None:
-            norm = np.linalg.norm(combined_embedding)
-            if norm > 0:
-                combined_embedding = combined_embedding / norm
+            # Process text queries
+            if text_queries and len(text_queries) > 0:
+                for query, weight in zip(text_queries, text_weights):
+                    query_embedding = self.embedding_service.encode_text_query(query)
 
-        # Build filter
-        filter_expr = self._build_rmag_filter(rmag_min, rmag_max)
+                    # Apply weight
+                    weighted_embedding = query_embedding * weight
 
-        # Search Zilliz
-        return self.zilliz_service.search(combined_embedding, top_k, filter_expr)
+                    if combined_embedding is None:
+                        combined_embedding = weighted_embedding
+                    else:
+                        combined_embedding += weighted_embedding
+
+            # Process image queries
+            if image_queries and len(image_queries) > 0:
+                if self.image_service is None:
+                    raise RuntimeError("Image service not initialized")
+
+                for img_query, weight in zip(image_queries, image_weights):
+                    # Encode image
+                    image_embedding = self.image_service.encode_image(
+                        ra=img_query['ra'],
+                        dec=img_query['dec'],
+                        fov=img_query.get('fov', 0.025),
+                        size=256
+                    )
+
+                    # Apply weight
+                    weighted_embedding = image_embedding * weight
+
+                    if combined_embedding is None:
+                        combined_embedding = weighted_embedding
+                    else:
+                        combined_embedding += weighted_embedding
+
+            # Normalize the final combined embedding
+            if combined_embedding is not None:
+                norm = np.linalg.norm(combined_embedding)
+                if norm > 0:
+                    combined_embedding = combined_embedding / norm
+
+            # Build filter
+            filter_expr = self._build_rmag_filter(rmag_min, rmag_max)
+
+            # Search Zilliz
+            return self.zilliz_service.search(combined_embedding, top_k, filter_expr)
+        except ValueError as e:
+            # Content moderation triggered - return empty results silently
+            if "moderation" in str(e).lower():
+                logger.info("Search blocked by content moderation")
+                return pd.DataFrame()
+            raise