I'm a 24-year-old Electrical and Computer Science Engineer currently working as a Research Associate at Carnegie Mellon University. I am also attending classes in the Master's program in Computer Science at CMU.
My research focuses on real-time differentiable rendering, 3D Gaussian Splatting, and computer vision, with deep specialization in high-performance computing and advanced graphics programming. I specialize in translating cutting-edge computer vision, machine learning, and graphics research into real-time functioning systems. I leverage CUDA optimization, Vulkan/OpenGL development, and advanced system architecture to push the boundaries of real-time rendering and neural scene reconstruction.
I have 3+ years of experience building production-ready systems that combine computer vision, 3D graphics, machine learning, and differentiable rendering. I'm passionate about bridging the gap between theoretical research and practical real-time applications, creating systems that make advanced graphics and ML techniques accessible for real-world deployment.
Email: alejandroamatp@gmail.com
Academic: aamat@andrew.cmu.edu
Location: Pittsburgh, PA
Phone: +1 412 512 9640
Research Interests
Professional Experience
Conducting research at the intersection of machine learning, computer graphics, and high-performance computing, specializing in real-time systems and optimization for 3D computer vision applications. Working in lab focused on realistic 3D reconstruction. Responsible for developing and optimizing cutting-edge rendering pipelines, GPU acceleration systems, and neural scene reconstruction algorithms.
- Contributed to static rendering component for accepted SIGGRAPH 2025 journal paper on 3D live streaming using Gaussian Splatting, achieving 3-4× speedup through load balancing kernels and CUDA graph pipelines with 15ms/iteration for 1.5M Gaussians
- Built high-performance computer vision systems with PyTorch prototyping and translation to real-time ML/graphics applications using advanced C++ optimizations including SIMD vectorization, OpenMP parallelization, Eigen library for fast matrix operations achieving 2-6ms/frame performance with cross-platform deployment using CMake and Python bindings
- Developed differentiable rendering pipeline using Slang/gfx API with custom GPU Adam optimizer and CUDA complete optimizations featuring automatic differentiation from fragment shaders, delivering 116× speedup over NVIDIA Falcor and 4× speedup in 2D Gaussian Splatting with profiling using Nsight Systems, Nsight Compute, and RenderDoc
- Train MLP architectures with PyTorch prototyping and TensorRT deployment, and develop custom differentiable optimizations with second-order nonlinear optimization algorithms across distributed systems
- Supervise visiting researchers and interns in 3D Gaussian Splatting and computer vision projects, conducting technical interviews and project coordination
- Pioneered real-time 3D imaging algorithms with MIMO FMCW mmWave radar for reflective surface optimization
- Published 2 papers on location-based utilization of reconfigurable intelligent surfaces for mmWave sensing
- Developed cross-platform VR application for IESE/MIT/LBS collaboration
- Deployed to Android, iOS, and Web/Desktop using Unity and Google Cardboard SDK
- Implemented Firebase integration for real-time data synchronization
- Led university team on collaborative project with HP, developing Python package using Intel RealSense cameras
- Implemented DBSCAN and RANSAC algorithms for pointcloud segmentation and wall plane detection
- Built data pipelines using Jenkins CI/CD, Airflow, MySQL, AWS, Python & Docker for retail analytics
- Developed automated data processing workflows for large-scale retail data analysis
- Implemented monitoring and logging systems for production data pipelines
Publications
"Echoes of the Coliseum: Towards 3D Live Streaming of Sports Events."
ACM Transactions on Graphics (SIGGRAPH 2025)
"Location-based real-time utilization of reconfigurable intelligent surfaces for mmWave integrated communication and sensing in full-immersive multiuser Metaverse scenarios."
IET Advanced Metaverse Wireless Communication Systems, pp. 101-136, 2025
Selected Projects
Cross-platform C++17 neural rendering engine breaking CUDA vendor lock-in for cross-platform neural scene rendering with Vulkan and GLSL compute shaders.
- Built complete 3DGS implementation with tile-based rasterization and multi-stage radix sort (histogram computation, work-efficient prefix sum using up-sweep/down-sweep phases across ping-pong buffers)
- Created ImGui keyframe animation system for real-time smooth sequence generation with interpolated camera movements, FOV control, culling, wireframe rendering and performance monitoring
- Achieved stable 60 fps on Nvidia 3060 Ti and 24-45 fps on Apple Silicon (MoltenVk translation) on 2M Gaussian pointcloud with bilinear upsampling for Retina display support
Built real-time differentiable rendering pipelines from scratch using Slang shader language, C++, and gfx API for inverse rendering applications.
- Developed inverse texturing system achieving 116× speedup over NVIDIA Falcor through automatic differentiation and optimized shader compilation
- Implemented 2D Gaussian Splatting optimization with 4× performance boost enabling real-time parameter optimization and scene reconstruction
- Created complete automatic differentiation framework with gradient computation for lighting, materials, and geometry parameters
Collection of GPU-accelerated algorithms with massive performance improvements through advanced CUDA optimization techniques.
- Sobel Edge Detection: 224,733× speedup over NumPy achieving 10.8 billion pixels/second throughput via shared memory tiling, warp optimization, and branch divergence elimination
- 3×3 SVD Implementation: 120× speedup vs PyTorch for computer graphics applications with custom CUDA kernels and Python bindings
- Fused MS-SSIM: Optimized Multi-Scale Structural Similarity Index implementation with PyTorch integration for real-time image quality assessment
Cross-platform VR application developed for prestigious business school collaboration between IESE, MIT, and London Business School.
- Built immersive educational experience using Unity and Google Cardboard SDK deployed across Android, iOS, and Web/Desktop platforms
- Implemented Firebase real-time synchronization for multi-user collaborative sessions and progress tracking
- Created intuitive VR interfaces optimized for educational content delivery and student engagement
Multiple custom Vulkan/OpenGL implementations featuring 3D rendering systems with advanced lighting models, mesh rendering, shadow mapping, camera controls, and optimized shader pipelines.
Custom x86 operating system kernel built from scratch in C and assembly language for educational and research purposes.
- Implemented complete process scheduling system with round-robin and priority-based algorithms, context switching, and multi-tasking support
- Built memory management subsystem featuring virtual memory, paging, page fault handling, and kernel/user space separation
- Developed system call interface, hardware interrupt handling, keyboard/timer drivers, and basic file system operations
Education
Currently attending classes in the MSCS program at CMU.
Thesis: "Real-Time Differentiable Rendering with Slang" - Built differentiable rendering pipelines achieving significant performance improvements. Supervised by Dr. Fernando de la Torre and Francisco Vicente at the Human Sensing Lab. Grade: 10/10
Electrical Engineering & Computer Science
GPA: 8.43/10 (Rank: 6/237) & 8.1/10 (Top 5%)
Award: 6th best student of the 2024 class out of 240 students
10 Honours in subjects including Audio Processing, Mobile Communications, AI, and Computer Graphics