Ryan Beldessi

Originally hired on as the UI/UX programmer in 2021 my role quickly evolved. Ultimately I was responsible for our technical design and implementation of systems shared between our wide range of products. I often was running out ahead of the other development teams in order to support their future work. This meant refactoring and rewriting systems; doing engine upgrades; building tooling for work that would be reoccurring; optimizing content for the Quest, Quest 2, and Quest 3; and formalizing the use of design patterns within the engineering teams. In addition to my work on our core technologies I was also the technical lead on our BodyWorks product (formerly Anatomy).

BodyWorks

I was the lead engineer on BodyWorks, a product for teaching anatomy. The project started as a way for the art and engineering team to do R&D on creating an interactive patient on headset that included all of the human anatomy. On this project I constructed tools to automate the generation of patients, these tools became the foundation for what would become our universal patient builder. I also worked with the art team to optimize geometry and shaders, while optimizing actor performance; these optimizations trickled down into all of the company’s other projects.

Character, Camera, Controls

When I started at Precision all aspects of the player were written exclusively in Blueprint. This was hugely restrictive to how engineers could approach building their products; everyone was locked in to working primarily in Blueprint. The required use of Blueprint was a huge source of frustration in the engineering team and leading to turnover. Nobody wanted ownership of the player because it was critical and incredibly frail. I stepped up and took ownership of it regardless.

Early on at Precision I refactored the player controller, pawn, avatar and hands to be exposed to the C++ via a UInterface (C++ interface with Blueprint support). This was a good stop-gap measure for immediately allowing the existing player systems to be referenced in C++.

Later in order to improve performance, reliability, and extensibility I rewrote the player systems from scratch in C++. My rewrite would allow spawning different pawns based on which headset was being used; this allowed us to more easily support future headsets (Pico, Apple Vision Pro) and separate Quest 3 exclusive feature implementation from the Quest 2. By rewriting the pawn in C++ I was also able to consolidate them down to a single actor which made spawning pawns in multiplayer faster and eliminated crashes related to player spawning during poor network conditions.

UI/UX

When starting at Precision the intention was I would be inheriting and extending upon an already existing UI. The UI I inherited used skeletal meshes to render what were essentially just textured quads, textured quads for images, and text renderers for text. There was no ability to align UI elements, text didn’t automatically line wrap, the UI was incredibly expensive to render because of the heavy use of skeletal meshes. It was immediately apparent all the UI needed a rebuild.

The first obvious step was moving the UI into Unreal’s built in UI systems. The core UI was rebuilt using widget blueprints and widget components. Immediately there was an obvious performance benefit but there was still a lot more that could be squeezed out. There was also an obvious decrease in text fidelity and generally significant aliasing on UI at distance. Besides visual fidelity there were also two obvious workflow issues introduced; it was difficult to align elements in the world with the UI, and it was difficult to trigger a redraw of the UI.

I derived classes from WidgetComponent and UserWidget to add additional features to simplify aligning components in world space to elements inside widgets, and to simplify triggering redraws of widgets from inside the widget itself.

Widget components do not generate and use mipmaps in Unreal. By overriding the UpdateRenderTarget function on PrecisionWidgetComponents I was able to add steps that would conditionally generate or clear mipmaps for the in use render target based on the widget’s state. Generating mipmaps on large widgets used a large chunk of the rendering budget (2-3ms). It would tank frame rates to generate mipmaps on all widget components on the same frame so it was necessary to very sparingly generate them on automatic redraws.

Localisation

While at PrecisionOS I constructed a localisation system that was highly portable and easy for our third-party partners to use. Although Unreal Engine has its own localisation system it was not built with sharing localisation data between multiple projects simultaneously in mind and the use of Portable Object files were too technical for our non-technical partners and employees. The localisation system consisted of three distinct parts. A localisation import tool, a localisation editor tool, and a localisation manager.

The localisation import tool was a C# console application that would process localisation data stored in Excel Spreadsheets and generate CSV files that were valid for use as string tables in Unreal Engine. Each string would need to be sanitized and made valid. This meant stripping command characters, inserting escape characters, etc. since users were largely non-technical let alone familiar with Unreal Engine or Unicode.

The localisation editor tool was an interface for editor users to modify localisation settings for their projects, view the status of localisation assets for their projects, quickly open localisation data assets, as well as import the most recent localisations. The editor tool would automate most asset actions like checking files out of Perforce, checking for Perforce locks, etc.

The localisation manager was responsible for setting the loaded language at edit and runtimes, and keeping language settings synced between clients in multiplayer. It needed to be available at edit time to support previewing languages in the widget editor. Essentially it worked by unregistering and registering CSV files as string tables based on what language was selected.