Below is a selection of my previous projects spanning research, teaching, and engineering. Each project is summarized briefly, with optional show details with some occasional behind-the-scenes context if you're curious to read more.
Tech Transfer
Pose in Adobe Fresco
Shipped in Adobe Fresco 7.0, 2025.
simulation, geometry, product development
In collaboration with the Drawing & Painting team and Adobe Research.After rejoining Adobe full-time, I joined early efforts on an intuitive, locality-aware, physics-based deformation tool for artists, which we shipped as Pose in Adobe Fresco in late 2025.
The neatest part is that it just works on any raster or vector layer, without any rigging, setup, or manual preparation.
By the way, Fresco is completely free and genuinely a fun app for painting and drawing! :)
Publications
An Adjoint Method for Differentiable Fluid Simulation on Flow Maps
ACM SIGGRAPH Asia 2025 (Conference Track)
simulation, physics-based machine learning
Zhiqi Li, Jinjin He, Barnabás Börcsök, Taiyuan Zhang, Duowen Chen, Tao Du, Ming C. Lin, Greg Turk, Bo ZhuDuring my time at Georgia Tech, I collaborated on a second fluid simulation paper exploring differentiable flow-map-based fluid solvers through an adjoint formulation.
The key strength of this approach is its ability to preserve vortical structures while enabling efficient gradient computation over long simulation horizons.
We demonstrate the method on a range of optimization tasks, including inverse problems (“what initial state produces a desired final outcome?”) and keyframe-based control, where external forces are optimized such that target density fields are matched at specified frames. The result on the left illustrates one such keyframe-matching scenario.
Lagrangian Covector Fluid with Free Surface
ACM SIGGRAPH 2024 (Conference Track)
simulation
Zhiqi Li, Barnabás Börcsök, Duowen Chen, Yutong Sun, Bo Zhu, Greg TurkOur paper explores free surface handling in particle-based (Lagrangian) covector fluid simulation. The method combines the advantages of particle-based and grid-based (Eulerian) approaches to fluid simulation, enabling more accurate and efficient simulation of complex fluid phenomena.
Controlling Laplacian Eigenfluids
BSc Thesis
simulation, physics-based machine learningIn my bachelor’s thesis , I explored controlling reduced-order fluid simulations using differentiable physics. Specifically, I investigated how neural networks can be trained to approximate control forces in a “Laplacian Eigenfluids” simulation by backpropagating through the simulation timesteps.
I found that solver-in-the-loop training quickly becomes computationally expensive for many practical scenarios. To address this, I focused on reduced-order representations with an interpretable latent space, spanned by eigenfunctions of the Laplace operator, enabling more efficient optimization and learning-based control. (You can see a couple of these Laplacian basis functions visualized on the left.)
My thesis was an extended version of my submission to the 2022 Scientific Students’ Associations’ Conference (TDK), where it received 1st prize with special distinction, and qualified for the National Conference (OTDK), receiving a Special Award in April 2023. I also presented the work at CESCG, where it received the 3rd Best Presentation Award.
Teaching
Differentiable Physics & Neural Networks
Guest Lecture for Computer Graphics in AI Era at Georgia Tech (Prof. Bo Zhu)
computer graphics, teaching, physics-based machine learningI had an opportunity to present a second guest lecture for Prof. Bo Zhu’s “Computer Graphics in AI Era” course at Georgia Tech, focusing on the intersection of physics and machine learning.
Physics in deep learning is a broad and fast-moving area, with so many exciting directions to explore. Drawing on ideas from sparse identification of non-linear dynamics (SINDy) and physics-informed neural networks (PINNs) to differentiable physics and hybrid models, my goal was to carve out a self-contained narrative that highlights when and why physical structure should be built directly into learning algorithms.
It was a lot of fun to distill these ideas into a form that encouraged discussion and curiosity, while remaining grounded in concrete examples from graphics and simulation.
3D Gaussian Splatting
Guest Lecture for Computer Graphics in AI Era at Georgia Tech (Prof. Bo Zhu)
computer graphics, teaching, computer vision, rendering, machine learningI had the opportunity to present a guest lecture on 3D Gaussian Splatting for Prof. Bo Zhu’s “Computer Graphics in AI Era” course at Georgia Tech, introducing point-based representations and their role in modern rendering and reconstruction pipelines.
With an audience ranging from undergraduates to experienced master’s students, I aimed to make the topic accessible without losing technical depth, starting from the underlying representation and optimization problem, and building up to practical system-level considerations.
It was a joy to present to such an enthusiastic group of students, and to see the topic spark questions across graphics, vision, and learning.
Volumetric Rendering
Guest Lecture for 3D Graphics at TU Budapest (Prof. László Szécsi)
computer graphics, teaching, renderingA guest lecture on volumetric rendering techniques, covering both foundational concepts and modern approaches for rendering participating media such as smoke, fog, clouds, and fire.
The lecture focused on building intuition for light transport in volumes, connecting physical principles with algorithmic choices used in practical renderers. Delivered as part of the “3D Graphics” course at TU Budapest in Spring 2023.
Design Principles Workshop
teaching, design, artI designed and presented a “Design Principles” workshop for our College of Advanced Studies, focusing on visual communication, layout, and fundamental design intuition.
I initially delivered the workshop in Spring 2021 and later revisited it in 2023. Recordings of both iterations are available online. (Content in Hungarian.)
Beginner's Web Development Workshop
teaching, web developmentA two-part beginner’s workshop introducing the basics of web development, covering HTML, CSS, and simple interactive components.
The sessions were designed to lower the barrier of entry for students with no prior experience, emphasizing hands-on experimentation and practical understanding. Source code and recordings are available through this git repository. (Content in Hungarian.)
Other Projects
Hand Gesture Detection for Interactive Physics Simulation
Georgia Tech (Group Project)
computer vision, image processing, machine learningA group project for Georgia Tech’s Computer Vision course, exploring how vision-based perception can be coupled with simulation to enable intuitive, human-in-the-loop interaction.
We connected MediaPipe ’s real-time hand gesture recognition to a live webcam feed, allowing hand gestures to directly control a real-time physics simulation. Clenching your fist spawns more smoke, while bending different fingers changes gravity. I’ve never felt closer to being an airbender!
The project explored latency, robustness, and mapping noisy perceptual signals to stable simulation controls. My main contributions focused on setting up MediaPipe, and implementing the fluid simulation.
Automatic Number Plate Recognition
TU Budapest (Group Project)
computer vision, image processing, machine learning, Python, PyTorch, OpenCVA group homework competition project for the Image Processing course at TU Budapest (1st prize).
We implemented an automatic number plate recognition pipeline using YOLOv7 for license plate detection and PaddleOCR for character recognition. The project focused on building a robust end-to-end system, handling variations in viewpoint, lighting, and image quality.
Differentiable Physics
TU Munich (Seminar Presentation)
simulation, physics-based machine learning, LaTeXA seminar presentation and short report on differentiable physics, centered around a recent research paper and the broader landscape of physics-based learning methods.
Written at TUM, during the Summer 2022 semester. (More details here.)
Rendering Participating Media
TU Munich (Seminar Presentation)
rendering, LaTeXAn overview of techniques for rendering participating media such as smoke, fog, clouds, and fire, with a focus on volumetric light transport and numerical integration.
Written at TUM, during the Winter 2021/2022 semester.
Update: this work later informed my guest lecture on volumetric rendering delivered at TU Budapest in Spring 2023.
Image-based Sports Detection
AIT-Budapest (Group Project)
deep learning, Python, Keras, OpenCVA group university assignment exploring image-based classification using deep learning.
We trained a neural network to identify the sport being played in an image, focusing on dataset construction, model selection, and evaluation. This project served as an early hands-on introduction to applied deep learning in computer vision.
Fluid and Cloth Simulation
TU Budapest (Individual Project)
simulation, C++, OpenGLAn individual project implementing basic fluid and cloth simulation methods using Smoothed-Particle Hydrodynamics (SPH) and Position-Based Dynamics (PBD).
The project focused on understanding core simulation principles, numerical stability, and real-time visualization using C++ and OpenGL.
Simulation of Curly Hair
TU Budapest (Individual Project)
simulation, C++, OpenGLA study and implementation of hair simulation techniques, culminating in a basic curly hair simulation system, implementing the Position-Based Dynamics (PBD) method.
This project marked one of my earliest deep dives into physically based animation and simulation.
Interactive Voronoi Partitioner
TU Budapest (Individual Project)
C++, SDL2An interactive Voronoi diagram visualizer implemented while learning C++ and the SDL2 graphics library.
This project served as an early exploration of computational geometry and interactive graphics programming.
Web Development
This Website :)
Hugo, DevOps, HTML, CSS, NGINX, LinuxI wanted to create a minimalist, performant, accessible, and accessible personal website that stays out of the way of the content.
The site is statically generated using Hugo, allowing me to write in simple, human-readable formats like Markdown, while keeping the build and deployment pipeline lightweight and relatively robust. Over a few iterations, I settled on a design that prioritizes clarity, longevity, and low maintenance over visual complexity.
Always under construction.
Roadmap
web development, MkDocs, CI/CD, knowledge managementAs the web development mentor for Schönherz Design Studio (schdesign) , I initiated and built Roadmap, a statically generated, collaboratively editable online knowledge base.
The goal was to make it as easy as possible for team members to share tutorials, write-ups, and recordings, while also smoothing the onboarding process for new contributors. The system emphasized simple authoring, version control, and automated deployment to lower friction for community-driven knowledge sharing. (Content in Hungarian.)
Web-Browser Joystick
web development, UIAn experimental web-based controller built for a large-scale interactive installation, where the side of our dormitory was turned into a massive LED display.
For a game played live on one of the biggest buildings in the neighborhood, more than a thousand viewers could use their smartphones as controllers. I was part of the team that designed and implemented the web-browser joystick for the game project.