diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
index beef33e0..2b37c41b 100644
--- a/.github/workflows/main.yml
+++ b/.github/workflows/main.yml
@@ -44,7 +44,7 @@ jobs:
         with:
           mode: start
           github-token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
-          ec2-image-id: ami-079a0c9425d45e778
+          ec2-image-id: ami-00bea05327004f12a
           ec2-instance-type: t3.2xlarge
           subnet-id: subnet-0c4a81ef57cf2ebe3
           security-group-id: sg-01cb8c0cb0de3ba00
diff --git a/images/datascience-notebook/Dockerfile b/images/datascience-notebook/Dockerfile
index f172251e..932162c2 100644
--- a/images/datascience-notebook/Dockerfile
+++ b/images/datascience-notebook/Dockerfile
@@ -1,4 +1,4 @@
-# If you are building manually, pass PYTHON_VERSION/PY_VER_SHORT/JUPYTERHUB_VERSION with --build-arg
+# If you are building manually, pass PYTHON_VERSION/PY_VER_SHORT/JUPYTERHUB_VERSION with --build-arg 
 ARG PYTHON_VERSION=python-3.11.8
 ARG PY_VER_SHORT=3.11
 ARG JUPYTERHUB_VERSION=4.1.5
@@ -64,32 +64,35 @@ USER jovyan
 
 # Python/Mamba Deps
 ## Package versions
-ARG JUPYTERSERVER_VERSION=2.14.2 NBGRADER_VERSION=0.9.3 JUPYTERLAB_VERSION=4.2.4 NBCONVERT_VERSION=7.16.4 NOTEBOOK_VERSION=7.2.1 NBCLASSIC_VERSION=1.1.0
+ARG JUPYTERHUB_VERSION=4.1.5 JUPYTERSERVER_VERSION=2.14.2 NBGRADER_VERSION=0.9.3 JUPYTERLAB_VERSION=4.2.4 NBCONVERT_VERSION=7.16.4 NOTEBOOK_VERSION=7.2.1 NBCLASSIC_VERSION=1.1.0
 ARG PANDAS_VERSION=2.2.3 STATSMODELS_VERSION=0.14.4 BOTTLENECK_VERSION=1.4.2 NUMEXPR_VERSION=2.10.2
 
+# Install uv for faster package installations
+RUN pip3 install --no-cache-dir --upgrade uv
+
 # Install essential+datascience pip packages 
 ## mistune added for nbgrader issues
-RUN mamba install -c conda-forge pillow typing-extensions tzlocal appdirs gputil mock pytest umap-learn && \
-    mamba install -c conda-forge nltk statsmodels=$STATSMODELS_VERSION pandas=$PANDAS_VERSION mistune && \
-    mamba install -c conda-forge dpkt nose datascience pyarrow bottleneck=$BOTTLENECK_VERSION umap-learn numexpr=$NUMEXPR_VESION && \
+RUN uv pip install pillow typing-extensions tzlocal appdirs gputil mock pytest umap-learn --system && \
+    uv pip install nltk statsmodels==$STATSMODELS_VERSION pandas==$PANDAS_VERSION mistune --system && \
+    uv pip install dpkt nose datascience pyarrow bottleneck==$BOTTLENECK_VERSION umap-learn numexpr==$NUMEXPR_VERSION --system && \
     python -c 'import matplotlib.pyplot' && \
     fix-permissions $CONDA_DIR && \
     fix-permissions /home/$NB_USER && \
-    mamba clean --all
+    uv cache clean
 
 # Install jupyterlab+extensions
-RUN mamba install -c conda-forge jupyterhub=$JUPYTERHUB_VERSION jupyter_server=$JUPYTERSERVER_VERSION && \
-    mamba install -c conda-forge jupyterlab=$JUPYTERLAB_VERSION notebook=$NOTEBOOK_VERSION nbclassic=$NBCLASSIC_VERSION && \
-    # (TODO: Re-enable collab once RTC is fixed) mamba install -c conda-forge jupyterlab_rise jupyter_server_terminals jupyter-collaboration && \
-    mamba install -c conda-forge jupyterlab_rise jupyter_server_terminals && \
-    mamba install -c conda-forge jupyterlab-latex jupyterlab-git jupyterlab-fasta jupyterlab-geojson && \
-    mamba install -c conda-forge nbconvert=$NBCONVERT_VERSION nbgrader=$NBGRADER_VERSION && \
+RUN uv pip install jupyterhub==$JUPYTERHUB_VERSION jupyter_server==$JUPYTERSERVER_VERSION --system && \
+    uv pip install jupyterlab==$JUPYTERLAB_VERSION notebook==$NOTEBOOK_VERSION nbclassic==$NBCLASSIC_VERSION --system && \
+    # (TODO: Re-enable collab once RTC is fixed) uv pip install jupyterlab_rise jupyter_server_terminals jupyter-collaboration --system && \
+    uv pip install jupyterlab_rise jupyter_server_terminals --system && \
+    uv pip install jupyterlab-latex jupyterlab-git jupyterlab-fasta jupyterlab-geojson --system && \
+    uv pip install nbconvert==$NBCONVERT_VERSION nbgrader==$NBGRADER_VERSION --system && \
     fix-permissions $CONDA_DIR && \
     fix-permissions /home/$NB_USER && \
-    mamba clean --all
+    uv cache clean
 
 # Install R packages
-RUN mamba install -c conda-forge r-markdown r-covr r-git2r r-crosstalk r-dt -y && \
+RUN mamba install -c conda-forge r-markdown r-covr r-git2r r-crosstalk r-dt r-sf -y && \
     fix-permissions $CONDA_DIR && \
     fix-permissions /home/$NB_USER && \
     mamba clean -a -y
@@ -99,6 +102,9 @@ RUN /usr/share/datahub/scripts/install-python-all.sh && \
     chown -R jovyan:users /opt/conda/etc/jupyter/nbconfig && \
     chmod -R +r /opt/conda/etc/jupyter/nbconfig
 
+RUN mamba remove --force libsqlite && \
+    mamba install -y libsqlite==3.46.1
+
 # Disable the jupyterlab extensions manager
 RUN jupyter labextension disable @jupyterlab/extensionmanager-extension
 
diff --git a/images/scipy-ml-notebook/Dockerfile b/images/scipy-ml-notebook/Dockerfile
index 17ec040f..bfad60a6 100644
--- a/images/scipy-ml-notebook/Dockerfile
+++ b/images/scipy-ml-notebook/Dockerfile
@@ -53,34 +53,31 @@ USER jovyan
 # CUDA setup w/mamba
 ## TODO: Investigate this command, seems to duplicate cuda packages for nvidia (pypi + conda-forge). 
 # cuda-toolkit is a skeleton package on CUDA 12, unlike CUDA <= 11
-RUN mamba install -c "nvidia/label/cuda-12.6.0" cuda-nvcc \
-    cuda-toolkit=$CUDA_VERSION \
-    # For CUDA 11: cudatoolkit=$CUDA_VERSION \
-    cuda-version=$CUDA_VERSION \
-    nccl \
-    -y && \
+RUN uv pip install --extra-index-url https://pypi.nvidia.com nvidia-cuda-nvcc-cu12 nvidia-nccl-cu12 cuda-python --system && \
     fix-permissions $CONDA_DIR && \
     fix-permissions /home/$NB_USER && \
-    mamba clean -a -y
+    uv cache clean
 
 # Install scipy pip packages
 ## install protobuf to avoid weird base type error. seems like if we don't then it'll be installed twice.
 ## https://github.com/spesmilo/electrum/issues/7825
-## pip cache purge didnt work here for some reason.
-RUN pip install --no-cache-dir protobuf==$PROTOBUF_VERSION
+RUN uv pip install --no-cache-dir protobuf==$PROTOBUF_VERSION --system && \
+    fix-permissions $CONDA_DIR && \ 
+    fix-permissions /home/$NB_USER && \
+    uv cache clean
+
 ## cuda-python installed to have parity with tensorflow and cudnn
 ## Install pillow<7 due to dependency issue https://github.com/pytorch/vision/issues/1712
 ## tensorrt installed to fix not having libnvinfer that has caused tensorflow issues.
-RUN pip install opencv-contrib-python-headless \
-    opencv-python && \
+RUN uv pip install opencv-contrib-python-headless opencv-python --system && \
     fix-permissions $CONDA_DIR && \ 
     fix-permissions /home/$NB_USER && \
-    pip cache purge
+    uv cache clean
 
-RUN mamba install -c conda-forge pyqt pycocotools pillow scapy && \
+RUN uv pip install PyQt5 pycocotools pillow scapy --system && \
     fix-permissions $CONDA_DIR && \
     fix-permissions /home/$NB_USER && \
-    mamba clean --all
+    uv cache clean
 
 # Install CUDA/Torch/Tensorflow/Keras w/pip
 ## no purge required but no-cache-dir is used. pip purge will actually break the build here!
@@ -88,33 +85,19 @@ RUN mamba install -c conda-forge pyqt pycocotools pillow scapy && \
 ## Check tensorrt_env_vars.sh if you have to bump tensorrt!
 
 ## tf-keras is keras 2.x for higher versions of tensorflow that would normally require keras 3
-RUN pip install nvidia-cudnn-cu12==$CUDNN_VERSION torch==$TORCH_VERSION torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126 && \
-  pip install tensorflow==$TENSORFLOW_VERSION tensorflow-datasets tensorrt==$TENSORRT_VERSION keras==$KERAS_VERSION tf-keras==$TF_KERAS_VERSION && \
+RUN uv pip install nvidia-cudnn-cu12==$CUDNN_VERSION torch==$TORCH_VERSION torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126 --system && \
+  uv pip install tensorflow==$TENSORFLOW_VERSION tensorflow-datasets tensorrt==$TENSORRT_VERSION keras==$KERAS_VERSION tf-keras==$TF_KERAS_VERSION --system && \
   fix-permissions $CONDA_DIR && \
   fix-permissions /home/$NB_USER && \
-  mamba clean -a -y && \
-  pip cache purge
+  uv cache clean
 
-RUN pip install transformers datasets accelerate huggingface-hub timm && \
+RUN uv pip install transformers datasets accelerate huggingface-hub timm --system && \
   fix-permissions $CONDA_DIR && \
   fix-permissions /home/$NB_USER && \
-  mamba clean -a -y && \
-  pip cache purge
+  uv cache clean
 
 USER $NB_UID:$NB_GID
 ENV PATH=${PATH}:/usr/local/nvidia/bin:/opt/conda/bin
 
-# CUDA fixes for CONDA
-## Copy libdevice file to the required path
-RUN mkdir -p $CONDA_DIR/lib/nvvm/libdevice && \
-    cp $CONDA_DIR/nvvm/libdevice/libdevice.10.bc $CONDA_DIR/lib/nvvm/libdevice/
-    #CUDA 11: cp $CONDA_DIR/lib/libdevice.10.bc $CONDA_DIR/lib/nvvm/libdevice/
-
-# TensorRT fix for tensorflow
-## https://github.com/tensorflow/tensorflow/issues/61468 (could not find TensorRT)
-## This will most definitely have to be changed after 8.6.1...
-RUN ln -s /opt/conda/lib/python3.11/site-packages/tensorrt_libs/libnvinfer_plugin.so.8 /opt/conda/lib/python3.11/site-packages/tensorrt_libs/libnvinfer_plugin.so.$TENSORRT_VERSION && \
-    ln -s /opt/conda/lib/python3.11/site-packages/tensorrt_libs/libnvinfer.so.8 /opt/conda/lib/python3.11/site-packages/tensorrt_libs/libnvinfer.so.$TENSORRT_VERSION
-
 # Run datahub scripts
 RUN . /tmp/activate.sh
diff --git a/images/spec.yml b/images/spec.yml
index 6add41e5..c1b1eafc 100644
--- a/images/spec.yml
+++ b/images/spec.yml
@@ -22,10 +22,10 @@ images:
     integration_tests: false
     info_cmds: [PY_VER, PIP_LIST, CUDA_VERSION, CONDA_INFO, CONDA_LIST, APT_PKG_LIST]
     prune: false # comment if scipy-ml stops being the last container
-    #prepull: false #-- uncomment to disable prepulling behavior for scipy-ml. gives you space on machine in exchange for build time.
+    # prepull: false #-- uncomment to disable prepulling behavior for scipy-ml. gives you space on machine in exchange for build time.
 
 tag:
-  prefix: "2025.2"
+  prefix: "2025.3"
 
 all_info_cmds:
   PY_VER: