diff --git a/ci/build_wheel_cpp.sh b/ci/build_wheel_cpp.sh
index 0367842a8c..ca27717769 100755
--- a/ci/build_wheel_cpp.sh
+++ b/ci/build_wheel_cpp.sh
@@ -36,4 +36,6 @@ python -m auditwheel repair \
     -w ${package_dir}/final_dist \
     ${package_dir}/dist/*
 
+./ci/validate_wheel.sh ${package_dir} final_dist
+
 RAPIDS_PY_WHEEL_NAME="${package_name}_${RAPIDS_PY_CUDA_SUFFIX}" rapids-upload-wheels-to-s3 cpp "${package_dir}/final_dist"
diff --git a/ci/build_wheel_python.sh b/ci/build_wheel_python.sh
index 0c709f2fe6..088e8e8e8f 100755
--- a/ci/build_wheel_python.sh
+++ b/ci/build_wheel_python.sh
@@ -25,4 +25,6 @@ python -m auditwheel repair \
     -w ${package_dir}/final_dist \
     ${package_dir}/dist/*
 
+./ci/validate_wheel.sh ${package_dir} final_dist
+
 RAPIDS_PY_WHEEL_NAME="${package_name}_${RAPIDS_PY_CUDA_SUFFIX}" rapids-upload-wheels-to-s3 python ${package_dir}/final_dist
diff --git a/ci/validate_wheel.sh b/ci/validate_wheel.sh
new file mode 100755
index 0000000000..5910a5c59f
--- /dev/null
+++ b/ci/validate_wheel.sh
@@ -0,0 +1,21 @@
+#!/bin/bash
+# Copyright (c) 2024, NVIDIA CORPORATION.
+
+set -euo pipefail
+
+package_dir=$1
+wheel_dir_relative_path=$2
+
+cd "${package_dir}"
+
+rapids-logger "validate packages with 'pydistcheck'"
+
+pydistcheck \
+    --inspect \
+    "$(echo ${wheel_dir_relative_path}/*.whl)"
+
+rapids-logger "validate packages with 'twine'"
+
+twine check \
+    --strict \
+    "$(echo ${wheel_dir_relative_path}/*.whl)"
diff --git a/python/kvikio/kvikio/benchmarks/s3_io.py b/python/kvikio/kvikio/benchmarks/s3_io.py
index 7941462650..5e1846a1e5 100644
--- a/python/kvikio/kvikio/benchmarks/s3_io.py
+++ b/python/kvikio/kvikio/benchmarks/s3_io.py
@@ -99,9 +99,33 @@ def run() -> float:
         yield run()
 
 
+def run_cudf(args, kvikio_remote_io: bool):
+    import cudf
+
+    cudf.set_option("kvikio_remote_io", kvikio_remote_io)
+    url = f"s3://{args.bucket}/data"
+
+    # Upload data to S3 server
+    create_client_and_bucket()
+    data = cupy.random.rand(args.nelem).astype(args.dtype)
+    df = cudf.DataFrame({"a": data})
+    df.to_parquet(url)
+
+    def run() -> float:
+        t0 = time.perf_counter()
+        cudf.read_parquet(url)
+        t1 = time.perf_counter()
+        return t1 - t0
+
+    for _ in range(args.nruns):
+        yield run()
+
+
 API = {
     "cupy": partial(run_numpy_like, xp=cupy),
     "numpy": partial(run_numpy_like, xp=numpy),
+    "cudf-kvikio": partial(run_cudf, kvikio_remote_io=True),
+    "cudf-fsspec": partial(run_cudf, kvikio_remote_io=False),
 }
 
 
@@ -135,7 +159,7 @@ def main(args):
         def pprint_api_res(name, samples):
             samples = [args.nbytes / s for s in samples]  # Convert to throughput
             mean = statistics.harmonic_mean(samples) if len(samples) > 1 else samples[0]
-            ret = f"{api}-{name}".ljust(12)
+            ret = f"{api}-{name}".ljust(18)
             ret += f"| {format_bytes(mean).rjust(10)}/s".ljust(14)
             if len(samples) > 1:
                 stdev = statistics.stdev(samples) / mean * 100
diff --git a/python/kvikio/pyproject.toml b/python/kvikio/pyproject.toml
index b002569646..cafde598bd 100644
--- a/python/kvikio/pyproject.toml
+++ b/python/kvikio/pyproject.toml
@@ -139,6 +139,14 @@ provider = "scikit_build_core.metadata.regex"
 input = "kvikio/VERSION"
 regex = "(?P<value>.*)"
 
+[tool.pydistcheck]
+select = [
+    "distro-too-large-compressed",
+]
+
+# PyPI limit is 100 MiB, fail CI before we get too close to that
+max_allowed_size_compressed = '75M'
+
 [tool.pytest.ini_options]
 filterwarnings = [
     "error",
diff --git a/python/libkvikio/pyproject.toml b/python/libkvikio/pyproject.toml
index 3be7cbc0ae..9504cb3755 100644
--- a/python/libkvikio/pyproject.toml
+++ b/python/libkvikio/pyproject.toml
@@ -59,3 +59,11 @@ requires = [
 
 [project.entry-points."cmake.prefix"]
 libkvikio = "libkvikio"
+
+[tool.pydistcheck]
+select = [
+    "distro-too-large-compressed",
+]
+
+# PyPI limit is 100 MiB, fail CI before we get too close to that
+max_allowed_size_compressed = '75M'