diff --git a/examples/llama.android/llama/src/main/cpp/kleidi-llama.cpp b/examples/llama.android/llama/src/main/cpp/kleidi-llama.cpp
index 2f090da5c7..f3fe26f018 100644
--- a/examples/llama.android/llama/src/main/cpp/kleidi-llama.cpp
+++ b/examples/llama.android/llama/src/main/cpp/kleidi-llama.cpp
@@ -23,7 +23,7 @@ static std::string join(const std::vector<T> &values, const std::string &delim)
 /**
  * Logging utils
  */
-#define TAG "llama-android.cpp"
+#define TAG "kleidi-llama"
 #define LOGv(...) __android_log_print(ANDROID_LOG_VERBOSE,  TAG, __VA_ARGS__)
 #define LOGd(...) __android_log_print(ANDROID_LOG_DEBUG,    TAG, __VA_ARGS__)
 #define LOGi(...) __android_log_print(ANDROID_LOG_INFO,     TAG, __VA_ARGS__)
@@ -177,7 +177,7 @@ Java_android_llama_cpp_internal_InferenceEngineImpl_benchModel(JNIEnv *env, jobj
                                                       jint pl, jint nr) {
     auto *context = init_context(g_model, pp);
     if (!context) {
-        const auto err_msg = "Fail to init_context! Bench aborted.";
+        const auto *const err_msg = "Fail to init_context! Bench aborted.";
         LOGe(err_msg);
         return env->NewStringUTF(err_msg);
     }
diff --git a/examples/llama.android/llama/src/main/java/android/llama/cpp/internal/InferenceEngineImpl.kt b/examples/llama.android/llama/src/main/java/android/llama/cpp/internal/InferenceEngineImpl.kt
index 114f7e4d19..1c42406c4a 100644
--- a/examples/llama.android/llama/src/main/java/android/llama/cpp/internal/InferenceEngineImpl.kt
+++ b/examples/llama.android/llama/src/main/java/android/llama/cpp/internal/InferenceEngineImpl.kt
@@ -4,6 +4,7 @@ import android.content.Context
 import android.llama.cpp.InferenceEngine
 import android.llama.cpp.UnsupportedArchitectureException
 import android.util.Log
+import dalvik.annotation.optimization.FastNative
 import kotlinx.coroutines.CancellationException
 import kotlinx.coroutines.CoroutineScope
 import kotlinx.coroutines.Dispatchers
@@ -37,7 +38,7 @@ import java.io.IOException
  *
  * State transitions are managed automatically and validated at each operation.
  *
- * @see llama-android.cpp for the native implementation details
+ * @see kleidi-llama.cpp for the native implementation details
  */
 internal class InferenceEngineImpl private constructor(
     private val nativeLibDir: String
@@ -74,20 +75,36 @@ internal class InferenceEngineImpl private constructor(
 
     /**
      * JNI methods
-     * @see llama-android.cpp
+     * @see kleidi-llama.cpp
      */
+    @FastNative
     private external fun init(nativeLibDir: String)
+
+    @FastNative
     private external fun load(modelPath: String): Int
+
+    @FastNative
     private external fun prepare(): Int
 
+    @FastNative
     private external fun systemInfo(): String
+
+    @FastNative
     private external fun benchModel(pp: Int, tg: Int, pl: Int, nr: Int): String
 
+    @FastNative
     private external fun processSystemPrompt(systemPrompt: String): Int
+
+    @FastNative
     private external fun processUserPrompt(userPrompt: String, predictLength: Int): Int
+
+    @FastNative
     private external fun generateNextToken(): String?
 
+    @FastNative
     private external fun unload()
+
+    @FastNative
     private external fun shutdown()
 
     private val _state =