diff --git a/cpp/include/raft/sparse/selection/knn.cuh b/cpp/include/raft/sparse/selection/knn.cuh
index 93b2a83bbd..e327386d13 100644
--- a/cpp/include/raft/sparse/selection/knn.cuh
+++ b/cpp/include/raft/sparse/selection/knn.cuh
@@ -33,7 +33,7 @@
 #include <raft/sparse/csr.cuh>
 #include <raft/sparse/distance/distance.cuh>
 #include <raft/sparse/selection/selection.cuh>
-#include <raft/spatial/knn/detail/brute_force_knn.cuh>
+
 #include <raft/spatial/knn/knn.hpp>
 
 #include <raft/cudart_utils.h>
@@ -310,7 +310,7 @@ class sparse_knn_t {
                         stream);
 
     // combine merge buffers only if there's more than 1 partition to combine
-    raft::spatial::knn::detail::knn_merge_parts(
+    raft::spatial::knn::knn_merge_parts(
       merge_buffer_dists, merge_buffer_indices, out_dists, out_indices,
       query_batcher.batch_rows(), 2, k, stream, trans.data());
   }
diff --git a/cpp/include/raft/spatial/knn/ann.hpp b/cpp/include/raft/spatial/knn/ann.hpp
new file mode 100644
index 0000000000..77d7831b4a
--- /dev/null
+++ b/cpp/include/raft/spatial/knn/ann.hpp
@@ -0,0 +1,82 @@
+/*
+ * Copyright (c) 2020-2021, NVIDIA CORPORATION.
+ *
+ * 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.
+ */
+
+#pragma once
+
+#include "ann_common.h"
+#include "detail/ann_quantized_faiss.cuh"
+
+#include <faiss/gpu/GpuIndex.h>
+#include <faiss/gpu/StandardGpuResources.h>
+
+#include <raft/mr/device/allocator.hpp>
+#include <raft/mr/device/buffer.hpp>
+
+namespace raft {
+namespace spatial {
+namespace knn {
+
+using deviceAllocator = raft::mr::device::allocator;
+
+/**
+ * @brief Flat C++ API function to build an approximate nearest neighbors index
+ * from an index array and a set of parameters.
+ *
+ * @param[in] handle RAFT handle
+ * @param[out] index index to be built
+ * @param[in] params parametrization of the index to be built
+ * @param[in] metric distance metric to use. Euclidean (L2) is used by default
+ * @param[in] metricArg metric argument
+ * @param[in] index_array the index array to build the index with
+ * @param[in] n number of rows in the index array
+ * @param[in] D the dimensionality of the index array
+ */
+template <typename value_idx = int>
+inline void approx_knn_build_index(raft::handle_t &handle,
+                                   raft::spatial::knn::knnIndex *index,
+                                   knnIndexParam *params,
+                                   raft::distance::DistanceType metric,
+                                   float metricArg, float *index_array,
+                                   value_idx n, value_idx D) {
+  detail::approx_knn_build_index(handle, index, params, metric, metricArg,
+                                 index_array, n, D);
+}
+
+/**
+ * @brief Flat C++ API function to perform an approximate nearest neighbors
+ * search from previously built index and a query array
+ *
+ * @param[in] handle RAFT handle
+ * @param[out] distances distances of the nearest neighbors toward
+ *                       their query point
+ * @param[out] indices indices of the nearest neighbors
+ * @param[in] index index to perform a search with
+ * @param[in] k the number of nearest neighbors to search for
+ * @param[in] query_array the query to perform a search with
+ * @param[in] n number of rows in the query array
+ */
+template <typename value_idx = int>
+inline void approx_knn_search(raft::handle_t &handle, float *distances,
+                              int64_t *indices,
+                              raft::spatial::knn::knnIndex *index, value_idx k,
+                              float *query_array, value_idx n) {
+  detail::approx_knn_search(handle, distances, indices, index, k, query_array,
+                            n);
+}
+
+}  // namespace knn
+}  // namespace spatial
+}  // namespace raft
diff --git a/cpp/include/raft/spatial/knn/ann_common.h b/cpp/include/raft/spatial/knn/ann_common.h
new file mode 100644
index 0000000000..6a6c7751c2
--- /dev/null
+++ b/cpp/include/raft/spatial/knn/ann_common.h
@@ -0,0 +1,75 @@
+/*
+ * Copyright (c) 2020-2021, NVIDIA CORPORATION.
+ *
+ * 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.
+ */
+
+#pragma once
+
+#include <raft/linalg/distance_type.h>
+
+#include <faiss/gpu/GpuIndex.h>
+#include <faiss/gpu/StandardGpuResources.h>
+
+namespace raft {
+namespace spatial {
+namespace knn {
+
+struct knnIndex {
+  faiss::gpu::GpuIndex *index;
+  raft::distance::DistanceType metric;
+  float metricArg;
+
+  faiss::gpu::StandardGpuResources *gpu_res;
+  int device;
+  ~knnIndex() {
+    delete index;
+    delete gpu_res;
+  }
+};
+
+enum QuantizerType : unsigned int {
+  QT_8bit,
+  QT_4bit,
+  QT_8bit_uniform,
+  QT_4bit_uniform,
+  QT_fp16,
+  QT_8bit_direct,
+  QT_6bit
+};
+
+struct knnIndexParam {
+  virtual ~knnIndexParam() {}
+};
+
+struct IVFParam : knnIndexParam {
+  int nlist;
+  int nprobe;
+};
+
+struct IVFFlatParam : IVFParam {};
+
+struct IVFPQParam : IVFParam {
+  int M;
+  int n_bits;
+  bool usePrecomputedTables;
+};
+
+struct IVFSQParam : IVFParam {
+  QuantizerType qtype;
+  bool encodeResidual;
+};
+
+};  // namespace knn
+};  // namespace spatial
+};  // namespace raft
diff --git a/cpp/include/raft/spatial/knn/detail/ann_quantized_faiss.cuh b/cpp/include/raft/spatial/knn/detail/ann_quantized_faiss.cuh
new file mode 100644
index 0000000000..bb37fd03db
--- /dev/null
+++ b/cpp/include/raft/spatial/knn/detail/ann_quantized_faiss.cuh
@@ -0,0 +1,210 @@
+/*
+ * Copyright (c) 2021, NVIDIA CORPORATION.
+ *
+ * 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.
+ */
+
+#pragma once
+
+#include "../ann_common.h"
+
+#include <raft/cudart_utils.h>
+#include <raft/cuda_utils.cuh>
+
+#include <label/classlabels.cuh>
+#include <raft/distance/distance.cuh>
+
+#include <faiss/gpu/GpuDistance.h>
+#include <faiss/gpu/GpuIndexFlat.h>
+#include <faiss/gpu/GpuIndexIVFFlat.h>
+#include <faiss/gpu/GpuIndexIVFPQ.h>
+#include <faiss/gpu/GpuIndexIVFScalarQuantizer.h>
+#include <faiss/gpu/GpuResources.h>
+#include <faiss/gpu/StandardGpuResources.h>
+#include <faiss/utils/Heap.h>
+#include <faiss/gpu/utils/Limits.cuh>
+#include <faiss/gpu/utils/Select.cuh>
+#include <faiss/gpu/utils/Tensor.cuh>
+
+#include <thrust/device_vector.h>
+#include <thrust/iterator/transform_iterator.h>
+
+#include <raft/linalg/distance_type.h>
+
+#include <cuml/common/device_buffer.hpp>
+#include <cuml/neighbors/knn.hpp>
+#include <raft/mr/device/allocator.hpp>
+
+#include <iostream>
+#include <set>
+
+namespace raft {
+namespace spatial {
+namespace knn {
+namespace detail {
+
+inline faiss::ScalarQuantizer::QuantizerType build_faiss_qtype(
+  QuantizerType qtype) {
+  switch (qtype) {
+    case QuantizerType::QT_8bit:
+      return faiss::ScalarQuantizer::QuantizerType::QT_8bit;
+    case QuantizerType::QT_8bit_uniform:
+      return faiss::ScalarQuantizer::QuantizerType::QT_8bit_uniform;
+    case QuantizerType::QT_4bit_uniform:
+      return faiss::ScalarQuantizer::QuantizerType::QT_4bit_uniform;
+    case QuantizerType::QT_fp16:
+      return faiss::ScalarQuantizer::QuantizerType::QT_fp16;
+    case QuantizerType::QT_8bit_direct:
+      return faiss::ScalarQuantizer::QuantizerType::QT_8bit_direct;
+    case QuantizerType::QT_6bit:
+      return faiss::ScalarQuantizer::QuantizerType::QT_6bit;
+    default:
+      return (faiss::ScalarQuantizer::QuantizerType)qtype;
+  }
+}
+
+template <typename IntType = int>
+void approx_knn_ivfflat_build_index(knnIndex *index, IVFParam *params,
+                                    raft::distance::DistanceType metric,
+                                    IntType n, IntType D) {
+  faiss::gpu::GpuIndexIVFFlatConfig config;
+  config.device = index->device;
+  faiss::MetricType faiss_metric = build_faiss_metric(metric);
+  faiss::gpu::GpuIndexIVFFlat *faiss_index = new faiss::gpu::GpuIndexIVFFlat(
+    index->gpu_res, D, params->nlist, faiss_metric, config);
+  faiss_index->setNumProbes(params->nprobe);
+  index->index = faiss_index;
+}
+
+template <typename IntType = int>
+void approx_knn_ivfpq_build_index(knnIndex *index, IVFPQParam *params,
+                                  raft::distance::DistanceType metric,
+                                  IntType n, IntType D) {
+  faiss::gpu::GpuIndexIVFPQConfig config;
+  config.device = index->device;
+  config.usePrecomputedTables = params->usePrecomputedTables;
+  config.interleavedLayout = params->n_bits != 8;
+  faiss::MetricType faiss_metric = build_faiss_metric(metric);
+  faiss::gpu::GpuIndexIVFPQ *faiss_index =
+    new faiss::gpu::GpuIndexIVFPQ(index->gpu_res, D, params->nlist, params->M,
+                                  params->n_bits, faiss_metric, config);
+  faiss_index->setNumProbes(params->nprobe);
+  index->index = faiss_index;
+}
+
+template <typename IntType = int>
+void approx_knn_ivfsq_build_index(knnIndex *index, IVFSQParam *params,
+                                  raft::distance::DistanceType metric,
+                                  IntType n, IntType D) {
+  faiss::gpu::GpuIndexIVFScalarQuantizerConfig config;
+  config.device = index->device;
+  faiss::MetricType faiss_metric = build_faiss_metric(metric);
+  faiss::ScalarQuantizer::QuantizerType faiss_qtype =
+    build_faiss_qtype(params->qtype);
+  faiss::gpu::GpuIndexIVFScalarQuantizer *faiss_index =
+    new faiss::gpu::GpuIndexIVFScalarQuantizer(index->gpu_res, D, params->nlist,
+                                               faiss_qtype, faiss_metric,
+                                               params->encodeResidual);
+  faiss_index->setNumProbes(params->nprobe);
+  index->index = faiss_index;
+}
+
+template <typename IntType = int>
+void approx_knn_build_index(raft::handle_t &handle,
+                            raft::spatial::knn::knnIndex *index,
+                            raft::spatial::knn::knnIndexParam *params,
+                            raft::distance::DistanceType metric,
+                            float metricArg, float *index_array, IntType n,
+                            IntType D) {
+  int device;
+  CUDA_CHECK(cudaGetDevice(&device));
+
+  faiss::gpu::StandardGpuResources *gpu_res =
+    new faiss::gpu::StandardGpuResources();
+  gpu_res->noTempMemory();
+  gpu_res->setDefaultStream(device, handle.get_stream());
+  index->gpu_res = gpu_res;
+  index->device = device;
+  index->index = nullptr;
+  index->metric = metric;
+  index->metricArg = metricArg;
+
+  // perform preprocessing
+  // k set to 0 (unused during preprocessing / revertion)
+  std::unique_ptr<MetricProcessor<float>> query_metric_processor =
+    create_processor<float>(metric, n, D, 0, false, handle.get_stream(),
+                            handle.get_device_allocator());
+
+  query_metric_processor->preprocess(index_array);
+
+  if (dynamic_cast<IVFFlatParam *>(params)) {
+    IVFFlatParam *IVFFlat_param = dynamic_cast<IVFFlatParam *>(params);
+    approx_knn_ivfflat_build_index(index, IVFFlat_param, metric, n, D);
+    std::vector<float> h_index_array(n * D);
+    raft::update_host(h_index_array.data(), index_array, h_index_array.size(),
+                      handle.get_stream());
+    query_metric_processor->revert(index_array);
+    index->index->train(n, h_index_array.data());
+    index->index->add(n, h_index_array.data());
+  } else {
+    if (dynamic_cast<IVFPQParam *>(params)) {
+      IVFPQParam *IVFPQ_param = dynamic_cast<IVFPQParam *>(params);
+      approx_knn_ivfpq_build_index(index, IVFPQ_param, metric, n, D);
+    } else if (dynamic_cast<IVFSQParam *>(params)) {
+      IVFSQParam *IVFSQ_param = dynamic_cast<IVFSQParam *>(params);
+      approx_knn_ivfsq_build_index(index, IVFSQ_param, metric, n, D);
+    } else {
+      ASSERT(index->index, "KNN index could not be initialized");
+    }
+
+    index->index->train(n, index_array);
+    index->index->add(n, index_array);
+    query_metric_processor->revert(index_array);
+  }
+}
+
+template <typename IntType = int>
+void approx_knn_search(raft::handle_t &handle, float *distances,
+                       int64_t *indices, raft::spatial::knn::knnIndex *index,
+                       IntType k, float *query_array, IntType n) {
+  // perform preprocessing
+  std::unique_ptr<MetricProcessor<float>> query_metric_processor =
+    create_processor<float>(index->metric, n, index->index->d, k, false,
+                            handle.get_stream(), handle.get_device_allocator());
+
+  query_metric_processor->preprocess(query_array);
+  index->index->search(n, query_array, k, distances, indices);
+  query_metric_processor->revert(query_array);
+
+  // Perform necessary post-processing
+  if (index->metric == raft::distance::DistanceType::L2SqrtExpanded ||
+      index->metric == raft::distance::DistanceType::L2SqrtUnexpanded ||
+      index->metric == raft::distance::DistanceType::LpUnexpanded) {
+    /**
+  * post-processing
+  */
+    float p = 0.5;  // standard l2
+    if (index->metric == raft::distance::DistanceType::LpUnexpanded)
+      p = 1.0 / index->metricArg;
+    raft::linalg::unaryOp<float>(
+      distances, distances, n * k,
+      [p] __device__(float input) { return powf(input, p); },
+      handle.get_stream());
+  }
+  query_metric_processor->postprocess(distances);
+}
+
+}  // namespace detail
+}  // namespace knn
+}  // namespace spatial
+}  // namespace raft
\ No newline at end of file
diff --git a/cpp/include/raft/spatial/knn/detail/brute_force_knn.cuh b/cpp/include/raft/spatial/knn/detail/knn_brute_force_faiss.cuh
similarity index 100%
rename from cpp/include/raft/spatial/knn/detail/brute_force_knn.cuh
rename to cpp/include/raft/spatial/knn/detail/knn_brute_force_faiss.cuh
diff --git a/cpp/include/raft/spatial/knn/knn.hpp b/cpp/include/raft/spatial/knn/knn.hpp
index eae5a3c8cc..a3a1972c13 100644
--- a/cpp/include/raft/spatial/knn/knn.hpp
+++ b/cpp/include/raft/spatial/knn/knn.hpp
@@ -16,7 +16,7 @@
 
 #pragma once
 
-#include "detail/brute_force_knn.cuh"
+#include "detail/knn_brute_force_faiss.cuh"
 
 #include <raft/mr/device/allocator.hpp>
 #include <raft/mr/device/buffer.hpp>
@@ -27,6 +27,15 @@ namespace knn {
 
 using deviceAllocator = raft::mr::device::allocator;
 
+template <typename value_idx = int64_t, typename value_t = float>
+inline void knn_merge_parts(value_t *inK, value_idx *inV, value_t *outK,
+                            value_idx *outV, size_t n_samples, int n_parts,
+                            int k, cudaStream_t stream,
+                            value_idx *translations) {
+  detail::knn_merge_parts(inK, inV, outK, outV, n_samples, n_parts, k, stream,
+                          translations);
+}
+
 /**
  * @brief Flat C++ API function to perform a brute force knn on
  * a series of input arrays and combine the results into a single