feat: rag-embedding-ai-chat (#1)

Co-authored-by: Sharang Parnerkar <parnerkarsharang@gmail.com>
Reviewed-on: #1

compliance-agent/src/api/handlers/graph.rs

@@ -52,7 +52,7 @@ pub async fn get_graph(
         // so there is only one set of nodes/edges per repo.
         let filter = doc! { "repo_id": &repo_id };
 
-        let nodes: Vec<CodeNode> = match db.graph_nodes().find(filter.clone()).await {
+        let all_nodes: Vec<CodeNode> = match db.graph_nodes().find(filter.clone()).await {
             Ok(cursor) => collect_cursor_async(cursor).await,
             Err(_) => Vec::new(),
         };
@@ -60,6 +60,17 @@ pub async fn get_graph(
             Ok(cursor) => collect_cursor_async(cursor).await,
             Err(_) => Vec::new(),
         };
+
+        // Remove disconnected nodes (no edges) to keep the graph clean
+        let connected: std::collections::HashSet<&str> = edges
+            .iter()
+            .flat_map(|e| [e.source.as_str(), e.target.as_str()])
+            .collect();
+        let nodes = all_nodes
+            .into_iter()
+            .filter(|n| connected.contains(n.qualified_name.as_str()))
+            .collect();
+
         (nodes, edges)
     } else {
         (Vec::new(), Vec::new())
@@ -235,12 +246,7 @@ pub async fn get_file_content(
     // Cap at 10,000 lines
     let truncated: String = content.lines().take(10_000).collect::<Vec<_>>().join("\n");
 
-    let language = params
-        .path
-        .rsplit('.')
-        .next()
-        .unwrap_or("")
-        .to_string();
+    let language = params.path.rsplit('.').next().unwrap_or("").to_string();
 
     Ok(Json(ApiResponse {
         data: FileContent {