diff --git a/compiler/record-hessian/CMakeLists.txt b/compiler/record-hessian/CMakeLists.txt
new file mode 100644
index 00000000000..75281e6c8cb
--- /dev/null
+++ b/compiler/record-hessian/CMakeLists.txt
@@ -0,0 +1,36 @@
+file(GLOB_RECURSE SOURCES "src/*.cpp")
+file(GLOB_RECURSE TESTS "src/*.test.cpp")
+list(REMOVE_ITEM SOURCES ${TESTS})
+
+add_library(record-hessian STATIC ${SOURCES})
+
+target_include_directories(record-hessian PUBLIC include)
+target_include_directories(record-hessian PRIVATE src)
+
+target_link_libraries(record-hessian luci_import)
+target_link_libraries(record-hessian luci_env)
+target_link_libraries(record-hessian luci_export)
+target_link_libraries(record-hessian luci_interpreter)
+target_link_libraries(record-hessian luci_log)
+target_link_libraries(record-hessian dio_hdf5)
+
+install(TARGETS record-hessian DESTINATION lib)
+install(DIRECTORY include/ DESTINATION include
+        FILES_MATCHING PATTERN "*.h")
+
+if(NOT ENABLE_TEST)
+  return()
+endif(NOT ENABLE_TEST)
+
+nnas_find_package(GTest REQUIRED)
+
+GTest_AddTest(record_hessian_tests ${TESTS})
+target_include_directories(record_hessian_tests PRIVATE include)
+target_include_directories(record_hessian_tests PRIVATE src)
+target_link_libraries(record_hessian_tests luci_lang)
+target_link_libraries(record_hessian_tests luci_pass)
+target_link_libraries(record_hessian_tests loco)
+target_link_libraries(record_hessian_tests dio_hdf5)
+target_link_libraries(record_hessian_tests nncc_coverage)
+target_link_libraries(record_hessian_tests luci_interpreter)
+target_link_libraries(record_hessian_tests record-hessian)
diff --git a/compiler/record-hessian/README.md b/compiler/record-hessian/README.md
new file mode 100644
index 00000000000..49e6b2d9365
--- /dev/null
+++ b/compiler/record-hessian/README.md
@@ -0,0 +1,3 @@
+# record-hessian
+
+_record-hessian_ calculates hessian metrix of activations for quantization.
diff --git a/compiler/record-hessian/include/record-hessian/HessianComputer.h b/compiler/record-hessian/include/record-hessian/HessianComputer.h
new file mode 100644
index 00000000000..fc3cdebcb93
--- /dev/null
+++ b/compiler/record-hessian/include/record-hessian/HessianComputer.h
@@ -0,0 +1,62 @@
+/*
+ * Copyright (c) 2024 Samsung Electronics Co., Ltd. All Rights Reserved
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef __RECORD_HESSIAN_HESSIANCOMPUTER_H__
+#define __RECORD_HESSIAN_HESSIANCOMPUTER_H__
+
+#include "record-hessian/HessianVector.h"
+
+#include <luci/IR/CircleNode.h>
+#include <luci_interpreter/Interpreter.h>
+
+#include <memory>
+#include <vector>
+#include <unordered_map>
+
+namespace record_hessian
+{
+/**
+ * @brief Record approximated hessian matrix from
+ * GPTQ paper(https://arxiv.org/abs/2210.17323).
+ */
+using HessianMap = std::unordered_map<const luci::CircleNode *, std::vector<float>>;
+using HessianVectorMap = std::unordered_map<const luci::CircleNode *, HessianVector>;
+
+class HessianComputer
+{
+public:
+  // Record min/max of node
+  void recordHessian(const luci::CircleNode *node, const luci_interpreter::Tensor *input_tensor);
+
+  std::unique_ptr<HessianMap> getMap();
+
+private:
+  HessianVectorMap _hessian_map;
+  const luci_interpreter::Tensor *_input_tensor = nullptr;
+
+  void recordHessianForConv2D(const luci::CircleNode *node);
+
+  void recordHessianForFullyConnected(const luci::CircleNode *node);
+};
+
+void unfold(std::vector<float> &buf, uint32_t input_n, uint32_t input_h, uint32_t input_w,
+            uint32_t input_c, uint32_t stride_h, uint32_t stride_w, uint32_t dilation_h,
+            uint32_t dilation_w, uint32_t kernel_oc, uint32_t kernel_h, uint32_t kernel_w,
+            uint32_t kernel_ic);
+
+} // namespace record_hessian
+
+#endif // __RECORD_HESSIAN_HESSIANCOMPUTER_H__
diff --git a/compiler/record-hessian/include/record-hessian/HessianVector.h b/compiler/record-hessian/include/record-hessian/HessianVector.h
new file mode 100644
index 00000000000..400b477616f
--- /dev/null
+++ b/compiler/record-hessian/include/record-hessian/HessianVector.h
@@ -0,0 +1,60 @@
+/*
+ * Copyright (c) 2024 Samsung Electronics Co., Ltd. All Rights Reserved
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef __RECORD_HESSIAN_HESSIANVECTOR_H__
+#define __RECORD_HESSIAN_HESSIANVECTOR_H__
+
+#include <luci/IR/CircleNodes.h>
+
+#include <cstddef>
+#include <vector>
+
+namespace record_hessian
+{
+
+struct HessianVector
+{
+  std::vector<float> hessian;
+  size_t count;
+
+  HessianVector() : count(0) {}
+
+  void update(const std::vector<float> &new_hessian)
+  {
+    if (count == 0)
+    {
+      hessian.resize(new_hessian.size());
+    }
+    else if (hessian.size() != new_hessian.size())
+    {
+      hessian.resize(new_hessian.size());
+    }
+
+    size_t numel = new_hessian.size();
+    float alpha = 1.f / static_cast<float>(count + 1);
+
+    for (size_t i = 0; i < numel; ++i)
+    {
+      hessian[i] = (hessian[i] * count + new_hessian[i]) * alpha;
+    }
+
+    count++;
+  };
+};
+
+} // namespace record_hessian
+
+#endif // __RECORD_HESSIAN_HESSIANVECTOR_H__
diff --git a/compiler/record-hessian/requires.cmake b/compiler/record-hessian/requires.cmake
new file mode 100644
index 00000000000..bfba787368f
--- /dev/null
+++ b/compiler/record-hessian/requires.cmake
@@ -0,0 +1,3 @@
+require("luci")
+require("luci-interpreter")
+require("dio-hdf5")
diff --git a/compiler/record-hessian/src/HessianComputer.cpp b/compiler/record-hessian/src/HessianComputer.cpp
new file mode 100644
index 00000000000..6ae36cf0797
--- /dev/null
+++ b/compiler/record-hessian/src/HessianComputer.cpp
@@ -0,0 +1,218 @@
+/*
+ * Copyright (c) 2024 Samsung Electronics Co., Ltd. All Rights Reserved
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "record-hessian/HessianComputer.h"
+
+#include <luci/IR/CircleQuantParam.h>
+
+namespace record_hessian
+{
+
+/**
+ * @brief unfold the vector with NHWC shape, inherently acting in an in-place manner.
+ * @note (N, H, W, C) -> (N, L, K_h * K_w * C).
+ * See details(https://pytorch.org/docs/stable/generated/torch.nn.Unfold.html).
+ */
+void unfold(std::vector<float> &buf, uint32_t input_n, uint32_t input_h, uint32_t input_w,
+            uint32_t input_c, uint32_t stride_h, uint32_t stride_w, uint32_t dilation_h,
+            uint32_t dilation_w, uint32_t kernel_oc, uint32_t kernel_h, uint32_t kernel_w,
+            uint32_t kernel_ic)
+{
+  assert(input_n > 0 && input_h > 0 && input_w > 0 && input_c > 0);
+  assert(stride_h > 0 && stride_w > 0);
+  assert(kernel_oc > 0 && kernel_h > 0 && kernel_w > 0 && kernel_ic > 0);
+
+  if (input_c != kernel_ic)
+    throw std::runtime_error("RecordHessian: Input channels do not match kernel channels.");
+  int out_height = (input_h - dilation_h * (kernel_h - 1) - 1) / stride_h + 1;
+  int out_width = (input_w - dilation_w * (kernel_w - 1) - 1) / stride_w + 1;
+  int patch_size = kernel_h * kernel_w * kernel_ic;
+  std::vector<float> unfolded_buf(input_n * out_height * out_width * patch_size, 0.0f);
+
+  int index = 0;
+  int in_y, in_x;
+  for (int n = 0; n < input_n; ++n)
+  {
+    for (int y = 0; y < out_height; ++y)
+    {
+      for (int x = 0; x < out_width; ++x)
+      {
+        for (int in_c = 0; in_c < input_c; ++in_c)
+        {
+          for (int ky = 0; ky < kernel_h; ++ky)
+          {
+            for (int kx = 0; kx < kernel_w; ++kx)
+            {
+              in_y = y * stride_h + ky * dilation_h;
+              in_x = x * stride_w + kx * dilation_w;
+              if (in_y < input_h && in_x < input_w)
+              {
+                unfolded_buf[index] = buf[((n * input_h + in_y) * input_w + in_x) * input_c + in_c];
+              }
+              index++;
+            }
+          }
+        }
+      }
+    }
+  }
+
+  buf.swap(unfolded_buf);
+}
+
+void HessianComputer::recordHessianForFullyConnected(const luci::CircleNode *node)
+{
+  assert(_input_tensor->shape().num_dims() < 4);
+  assert(_input_tensor->element_type() == luci_interpreter::DataType::FLOAT32);
+
+  uint32_t size_in_ch;
+  uint32_t length;
+
+  const auto data = _input_tensor->data<float>();
+  const auto num_elements = _input_tensor->shape().num_elements();
+  std::vector<float> buf(data, data + num_elements);
+
+  if (_input_tensor->shape().num_dims() == 3)
+  {
+    size_in_ch = _input_tensor->shape().dim(2); // input_tensor [batch, length, channel]
+  }
+  else if (_input_tensor->shape().num_dims() == 2)
+  {
+    size_in_ch = _input_tensor->shape().dim(1); // input_tensor [length, channel]
+  }
+  else
+  {
+    throw std::runtime_error("RecordHessian: Unsupported node rank");
+  }
+  assert(size_in_ch != 0);
+  length = num_elements / size_in_ch;
+
+  std::vector<float> hessian(size_in_ch * size_in_ch, 0);
+
+  for (int i = 0; i < size_in_ch; ++i)
+  {
+    for (int j = 0; j < size_in_ch; ++j)
+    {
+      float sum = 0;
+      for (int k = 0; k < length; ++k)
+      {
+        sum += buf[i + k * size_in_ch] * buf[j + k * size_in_ch];
+      }
+      hessian[i * size_in_ch + j] = 2 * sum;
+    }
+  }
+
+  HessianVector &vector = _hessian_map[node];
+  vector.update(hessian);
+}
+
+void HessianComputer::recordHessianForConv2D(const luci::CircleNode *node)
+{
+  assert(_input_tensor->shape().num_dims() == 4);
+  assert(_input_tensor->element_type() == luci_interpreter::DataType::FLOAT32);
+
+  const auto circle_conv2d = loco::must_cast<const luci::CircleConv2D *>(node);
+  const auto node_filter = loco::must_cast<luci::CircleConst *>((circle_conv2d)->filter());
+  assert(circle_conv2d->rank() >= 4);
+  assert(node_filter->dtype() == loco::DataType::FLOAT32);
+  assert(node_filter->rank() == 4);
+
+  uint32_t size_filter = node_filter->size<loco::DataType::FLOAT32>();
+  uint32_t size_in_ch =
+    node_filter->size<loco::DataType::FLOAT32>() / circle_conv2d->dim(3).value();
+
+  uint32_t input_n = _input_tensor->shape().dim(0);
+  uint32_t input_h = _input_tensor->shape().dim(1);
+  uint32_t input_w = _input_tensor->shape().dim(2);
+  uint32_t input_c = _input_tensor->shape().dim(3);
+
+  uint32_t stride_h = circle_conv2d->stride()->h();
+  uint32_t stride_w = circle_conv2d->stride()->w();
+  uint32_t dilation_h = circle_conv2d->dilation()->h();
+  uint32_t dilation_w = circle_conv2d->dilation()->w();
+
+  uint32_t kernel_oc = node_filter->dim(0).value();
+  uint32_t kernel_h = node_filter->dim(1).value();
+  uint32_t kernel_w = node_filter->dim(2).value();
+  uint32_t kernel_ic = node_filter->dim(3).value();
+
+  const auto data = _input_tensor->data<float>();
+  const auto num_elements = _input_tensor->shape().num_elements();
+  assert(data != 0);
+  assert(num_elements != 0);
+  std::vector<float> buf(data, data + num_elements);
+
+  unfold(buf, input_n, input_h, input_w, input_c, stride_h, stride_w, dilation_h, dilation_w,
+         kernel_oc, kernel_h, kernel_w, kernel_ic);
+  assert(size_in_ch != 0);
+  uint32_t length = buf.size() / size_in_ch;
+
+  std::vector<float> hessian(size_in_ch * size_in_ch, 0);
+  for (int i = 0; i < size_in_ch; ++i)
+  {
+    for (int j = 0; j < size_in_ch; ++j)
+    {
+      float sum = 0;
+      for (int k = 0; k < length; ++k)
+      {
+        sum += buf[i + k * size_in_ch] * buf[j + k * size_in_ch];
+      }
+      hessian[i * size_in_ch + j] = 2 * sum;
+    }
+  }
+
+  HessianVector &vector = _hessian_map[node];
+  vector.update(hessian);
+}
+
+void HessianComputer::recordHessian(const luci::CircleNode *node,
+                                    const luci_interpreter::Tensor *input_tensor)
+{
+  if (node == nullptr || input_tensor == nullptr)
+    throw std::invalid_argument("RecordHessian: node or input_tensor is null.");
+
+  if (input_tensor->element_type() != luci_interpreter::DataType::FLOAT32)
+    throw std::runtime_error("RecordHessian: Unsupported dtype: only FLOAT32 is supported.");
+
+  _input_tensor = input_tensor;
+
+  switch (node->opcode())
+  {
+    case luci::CircleOpcode::FULLY_CONNECTED:
+      recordHessianForFullyConnected(node);
+      break;
+    case luci::CircleOpcode::CONV_2D:
+      recordHessianForConv2D(node);
+      break;
+    default:
+      throw std::runtime_error("RecordHessian: " + node->name() + " is unsupported op.");
+  }
+}
+
+std::unique_ptr<HessianMap> HessianComputer::getMap()
+{
+  auto hessian_map = std::make_unique<HessianMap>();
+
+  for (auto item : _hessian_map)
+  {
+    auto &vec = (*hessian_map)[item.first];
+    vec = item.second.hessian;
+  }
+
+  return hessian_map;
+}
+
+} // namespace record_hessian
diff --git a/compiler/record-hessian/src/HessianComputer.test.cpp b/compiler/record-hessian/src/HessianComputer.test.cpp
new file mode 100644
index 00000000000..d64ab99678d
--- /dev/null
+++ b/compiler/record-hessian/src/HessianComputer.test.cpp
@@ -0,0 +1,108 @@
+/*
+ * Copyright (c) 2024 Samsung Electronics Co., Ltd. All Rights Reserved
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "record-hessian/HessianComputer.h"
+
+#include <luci/IR/CircleNode.h>
+#include <luci_interpreter/Interpreter.h>
+
+#include <gtest/gtest.h>
+
+#include <vector>
+
+using namespace record_hessian;
+
+TEST(HessianComputerTest, recordHessianValidInput)
+{
+  luci::CircleFullyConnected node;
+
+  std::vector<float> input_data = {1.0, 2.0, 3.0, 4.0};
+
+  luci_interpreter::DataType data_type = luci_interpreter::DataType::FLOAT32;
+  luci_interpreter::Shape shape({1, 4});
+  luci_interpreter::AffineQuantization quantization;
+
+  std::string tensor_name = "input_tensor";
+
+  luci_interpreter::Tensor input_tensor(data_type, shape, quantization, tensor_name);
+
+  size_t data_size = input_data.size() * sizeof(float);
+  std::vector<uint8_t> buffer(data_size);
+
+  input_tensor.set_data_buffer(buffer.data());
+  input_tensor.writeData(input_data.data(), data_size);
+
+  HessianComputer computer;
+
+  EXPECT_NO_THROW(computer.recordHessian(&node, &input_tensor));
+}
+
+TEST(HessianComputerTest, recordHessian_wrong_op_NEG)
+{
+  luci::CircleAdd node;
+
+  std::vector<float> input_data = {1.0, 2.0, 3.0, 4.0};
+
+  luci_interpreter::DataType data_type = luci_interpreter::DataType::FLOAT32;
+  luci_interpreter::Shape shape({1, 2, 2, 1});
+  luci_interpreter::AffineQuantization quantization;
+
+  std::string tensor_name = "input_tensor";
+
+  luci_interpreter::Tensor input_tensor(data_type, shape, quantization, tensor_name);
+
+  size_t data_size = input_data.size() * sizeof(float);
+  std::vector<uint8_t> buffer(data_size);
+
+  input_tensor.set_data_buffer(buffer.data());
+  input_tensor.writeData(input_data.data(), data_size);
+
+  HessianComputer computer;
+
+  EXPECT_ANY_THROW(computer.recordHessian(&node, &input_tensor));
+}
+
+TEST(HessianComputerTest, recordHessianNullTensor_NEG)
+{
+  luci::CircleAdd node;
+  HessianComputer computer;
+  EXPECT_ANY_THROW(computer.recordHessian(&node, nullptr));
+}
+
+TEST(HessianComputerTest, unfoldValidInput)
+{
+  std::vector<float> buf = {1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0};
+  uint32_t input_n = 1, input_h = 2, input_w = 2, input_c = 2;
+  uint32_t stride_h = 1, stride_w = 1, dilation_h = 1, dilation_w = 1;
+  uint32_t kernel_oc = 1, kernel_h = 2, kernel_w = 2, kernel_ic = 2;
+
+  unfold(buf, input_n, input_h, input_w, input_c, stride_h, stride_w, dilation_h, dilation_w,
+         kernel_oc, kernel_h, kernel_w, kernel_ic);
+  std::vector<float> expected_output = {1.0, 3.0, 5.0, 7.0, 2.0, 4.0, 6.0, 8.0};
+
+  EXPECT_EQ(buf, expected_output);
+}
+
+TEST(HessianComputerTest, unfoldInvalidInput_NEG)
+{
+  std::vector<float> buf = {1.0, 2.0, 3.0, 4.0};
+  uint32_t input_n = 1, input_h = 2, input_w = 2, input_c = 1;
+  uint32_t stride_h = 1, stride_w = 1, dilation_h = 1, dilation_w = 1;
+  uint32_t kernel_oc = 1, kernel_h = 2, kernel_w = 2, kernel_ic = 2;
+
+  EXPECT_ANY_THROW(unfold(buf, input_n, input_h, input_w, input_c, stride_h, stride_w, dilation_h,
+                          dilation_w, kernel_oc, kernel_h, kernel_w, kernel_ic));
+}