diff --git a/paper/paper.md b/paper/paper.md
index 9280d54..cd7a076 100644
--- a/paper/paper.md
+++ b/paper/paper.md
@@ -37,7 +37,7 @@ Without such a tool, design spaces cannot be efficiently explored to locate opim
 Commonly, geometry design and generation will take place using a computer-aided design (CAD) software.
 Although commercial CAD packages offer a comprehensive toolset for design, they are not usually intended for large-scale, automated design optimisation.
 As such, they do not have the functionality desired for such a purpose.
-There are also existing Python-native geometry packages [@numpy_stl, @PyMesh], which provide a good foundation for a more flexible application, as they offer a low-level API for the general purpose of generating meshes from vertex coordinate and face IDs.
+There are also existing Python-native geometry packages [@numpy_stl; @PyMesh], which provide a good foundation for a more flexible application, as they offer a low-level API for the general purpose of generating meshes from vertex coordinate and face IDs.
 HyperVehicle differs from and extends these packages in many ways.
 
 
@@ -47,10 +47,10 @@ This does not mean that the package cannot be used for more general geometry gen
 It also features all of the generic tools which are commonly found in a CAD software, but with the convenience and flexibility that Python affords, making it easy to integrate with.
 Importantly, HyperVehicle also offers the capability to differentiate geometry with respect to design parameters, making it an extremely useful component of a gradient-based design optimisation framework.
 This is exemplified in the hypersonic aerodynamics design and optimisation 
-papers relying on HyperVehicle [@MackleAFMC, @MackleShapeOpt, @MackleCoDesign].
+papers relying on HyperVehicle [@MackleAFMC; @MackleShapeOpt; @MackleCoDesign].
 
 
-Similar frameworks appear abundantly throughout literature [@Bowcutt1992, @Bowcutt2001], yet there does not appear to be a publicly available library offering the required capabilities.
+Similar frameworks appear abundantly throughout literature [@Bowcutt1992; @Bowcutt2001], yet there does not appear to be a publicly available library offering the required capabilities.
 HyperVehicle aims to fill this gap, by providing an open-source, user-friendly and production-ready automated CAD tool as a Python package.
 
 
@@ -59,9 +59,9 @@ Below are some sample vehicle configurations generated using HyperVehicle.
 Note that these geometries are made avaiable in the hypervehicle hangar.
 More examples can be seen on the documentation website.
 
-![Generic hypersonic waverider](images/waverider.png){ width=70% }
-![DLR ReFEX](images/refex.png){ width=70% }
-![NASA X43-A demonstrator](images/x43a.png){ width=70% }
+![Generic hypersonic waverider](images/waverider.png){ width=33% }
+![DLR ReFEX](images/refex.png){ width=33% }
+![NASA X43-A demonstrator](images/x43a.png){ width=33% }
 
 
 # References