Hi, I'm Umberto.

I'm working on bringing High on Life 2 to Nintendo Switch 2 in a co-dev setup with Squanch Games. It's an ongoing project so there's a lot of communication with the original team through Slack. Issues, bugs and tasks are taken care of as development moves forward.

• Profiling CPU performance with Unreal Insights, focusing on physics, AI and animation costs
• Bug-fixing alongside the QA team, all tracked in Jira
• Using Nintendo Switch 2 remote SDK tools and CVars to pinpoint platform-specific issues
• Reading through platform-specific Unreal Engine code to get my analysis done

My first big porting project at Untold Games. High on Life is a bizarre FPS by Squanch Games and getting it to run well on Switch, and later Switch 2, was genuinely challenging. The hardware constraints forced me to really understand how Unreal manages game threads and tasks, which made me aware of the "anatomy" of the engine.

• Worked on different custom game systems, mainly regarding physics, AI and animation
• Used Unreal Insights and Nintendo's SDK tools to conduct detailed performance analysis
• Optimized several key systems to hit target framerate on Switch 1 & 2 hardware while preventing CPU spikes
• Trimmed CPU time to free up budget and stabilize framerate during gameplay
• Fixed a long list of platform-specific bugs across Switch 1 and Switch 2, tracking them with the QA team through Jira
• Dug into Unreal Engine's platform-specific code using CVars for debugging

Porting Police Simulator to Switch was interesting because it wasn't just a technical job, but involved technical UI design skills. The original PC/console interface didn't translate well to Switch's screen size and resolution, so I had to rethink how players would interact with it. I ended up doing proper competitor research for open-world games on Switch before writing the design documentation with the proposed solution.

• Designed and documented additional UI features for the Switch version
• Working alongside the Aesir team, I researched, presented, designed and iterated on my designs
• Profiled CPU performance and optimized bottlenecks using Unreal Insights and Nintendo SDK tools
• Debugged platform-specific issues using Nintendo's tools and console variable testing
• Bug fixing and QA tracking via Jira throughout the porting process

This was my first leading role experience. I pitched the concept during the university course, which was ranked among the top proposals (out of ~100) and selected for development; suddenly I was responsible for a team of 5.

You wake up as a mysterious child in a laboratory facility, but soon you discover that you can manipulate time by making it flow or stop. When time stops you can move but when time flows you can't. You will be able to channel forces onto objects to make them move and solve the tests that scientists planned for you...but nothing is what it seems and soon you will discover the secrets behind the laboratory.

• Came up with the concept, pitched it, and kept the design documentation updated throughout production
• Led the team: kept morale up, assigned tasks weekly and made sure we were on schedule
• Designed all the tutorial levels and 3 puzzles on paper first, then built every level from blockout to final
• Coded the camera system, character movement, force mechanic, post-processing and cutscenes on Unity in C#
• Tuned the difficulty curve to make sure the game taught itself without being frustrating, taking Portal's progression as the main inspiration
• Directed the artist with references and made sure the environment and music matched the tone I had in mind, building tension as the levels progress and reveal the final room

My MSc thesis, done in collaboration with AnotheReality. The idea was to build a VR co-op game for corporate team-building and a tool that could automatically generate maze instances at a target difficulty level. The difficulty was validated by simulating gameplay with AI agents that behaved like real player archetypes — cautious, aggressive, or smart. The research ended up getting published on IEEE.

• Designed the VR co-op game and its team-building mechanics from scratch
• Built the PCG tool that generates maze layouts at a controllable difficulty
• Implemented AI agents with distinct behaviour profiles to simulate and measure how hard each maze actually was
• Wrote the full thesis covering PCG theory, history and the specific algorithms I used
• Got the research published in an IEEE scientific journal

A hyper-casual mobile game made during the Game Design course at Digital Bros Game Academy. You're an arrow flying through Japanese-themed levels: dodge civilians, kill only ninjas, beat the boss. Simple premise, but mastering the enemy patterns and the boss fight is what allows you to score the maximum. We worked in Scrum with Trello, which was my first taste of a more structured production pipeline.

• Helped write the initial pitch and shape the concept with the team
• Designed and balanced the final boss fight
• Built the tutorial with a focus on teaching the game intuitively and with the least cognitive load for the player
• Implemented a tool to design enemy's animation patterns using AnimationCurve component in Unity
• Did a hyper-casual competitor analysis during pre-production to make sure we were aiming at the right target

Game jams are probably my favourite way to experiment with game design. No long production cycles, no scope creep: just a weekend to build something that works and feels good. I did my first jam in 2020, taking part in the Global Game Jam (GGJ) but only in the last two years have I been doing them frequently, taking part in 8 Game Jams including the 2026 GGJ. I proposed my intentions to my bosses who gave the space and resources at Untold Games's office, which turned out to be a great way to connect with colleagues outside of the usual work context and develop strange and entertaining games. Not all the jams are meant to turn out great but there's plenty of room to learn professionally, as a game developer, and personally. The ones you find at my itch.io page are the best outcomes and each one of them makes me learn something new.

• Organized and coordinated internal jam events at Untold Games
• Led the brainstorm from idea exploration to convergence, coming up with a one-pager defining the verbs and nouns of the game
• Handled task assignment and kept track of progress as Project Manager
• Designed and programmed across all jam projects

Internal Jam · Team of 5

Shut up and Kill me!

Multiplayer Party

Learned how important it is to work around dynamics. After brainstorming we discussed a lot about what the final game feels like for the player and the game ended up with good tension and meaningful strategic choices between players, as intended despite some possible minor fixes and balancing.

Play

Metroidvania Month 30 37th / 122 · Team of 2

Mætʃ

Top-down Metroidvania

Learned that small size teams are the best when the members empathize with each other. The jam theme was "minimalist" metroidvania and so, before brainstorming, we did a complete research on both the artistic movement and the genre, with a focus on action-adventure games. Thanks to our shared interests and tastes, the direction of the game came up very smoothly and it was appreciated by players, and so did we.

Play

Brackeys 2026.1 342nd / 1429 · Team of 8

CRT.exe

Puzzle Adventure

Learned the difficulties of larger teams, mostly due to the diverse individual views on many design aspects. At the end we managed to deliver the game on time with some cool ideas for level switching that make the game still an enjoyable and complete experience.

Play

I've always been curious about AI and machine learning, especially its application to videogames. I used UE4's advanced vehicle template as a starting point and imported the Unreal Engine Python plugin: it embeds a whole Python VM that can be used in editor and, most importantly, at runtime! After I imported it I had access to all the AI Python libraries and decided to use TensorFlow to train agents using neuroevolution: basically letting a population of cars evolve over generations until they could drive a full track without crashing. What I found most interesting was testing whether agents trained on harder tracks could generalise better to new tracks. As per the basic rules of Machine Learning, the train set should be the most exhaustive possible so that the best learned model can generalize better on all the possible tracks.

• Started from UE4 Advanced Vehicle template
• Imported a plugin to use Python with TensorFlow at runtime
• Neuroevolutionary training loop on three different tracks: population selection, mutation and fitness scoring each generation
• The best-performing agents of all three were tested on unseen irregular tracks to check generalisation
• Analyzed the results from simple vs complex training tracks and observed the overfitting behaviour of simpler ones

Training:

↓

Test:

Testing of the best models found during Neuroevolution training.
The agents found are tested on an unseen and irregular track: as expected, learning from simple tracks will make the car collide easily, while learning from complex tracks will make the car behave correctly.

I built this project to get comfortable with Behaviour Trees and Environment Query System in Unreal Engine. You can decide to spawn three kinds of NPCs, gradually more social, that live out a daily routine on their own: they pick activities, interact with each other when they feel like it, and go to sleep at night. No scripted sequences, all driven by the AI systems.

• Behaviour Trees driving all NPC decision-making and activity scheduling
• Environment Query System (EQS) for spatial awareness and activity location selection
• Red, Yellow and Blue NPCs that gradually favour cooperative activities
• NPCs decide whether to do a solo or coop activity, then go find the right place
• Coop activities are special since the initiator goes to search NPCs that want to join: NPCs decide to suspend their current activity depending on their type and their preference for cooperative tasks
• Solo activities: Idle, Sitting, Reading, Smoking, Drinking Coffee, Playing on a console
• Cooperative activities: Conversation, Ping Pong, Playing on a console together
• Night cycle: NPCs stop all activities and autonomously go to sleep