Hi, my name is Giorgia Carboni
I'm a Computer Vision Engineer passionate about AI, Machine Learning and Game Development.
Experience
Computer Vision R&D Engineer
Fine-tuned VLMs using QLoRA for industrial environments, reducing memory footprint by 2.5x for consumer GPUs. Built automated synthetic data pipelines for spatial positioning and object classification. Developed asynchronous microservices (FastAPI, Celery) for Mixed Reality assistants and orchestrated PyTorch frameworks for 2D-3D embedding alignment (CLIP/Sonata).
Engineer
Deployed and managed services (Tomcat, JBoss) on Linux VMs. Handled incident management and executed Windows Server migrations to ensure operational continuity.
〰
Projects I'm working on:
Master Thesis: High-Fidelity 3D Gaussian Splatting (Python/C++)
In Progress.
Developing a framework for High-Fidelity generation from SfM (COLMAP) on consumer hardware.
Focusing on kernel optimization for real-time rasterization, volumetric rendering efficiency, and temporal super-sampling for dynamic scenes.
GPU-Based Ray-Casting of Quadratic Surfaces (C++/OpenGL/GLSL)
In Progress.
Collaborative academic project implementing per-pixel ray-casting for quadratic primitives via a programmable GPU pipeline. Bypasses standard tessellation by solving homogeneous intersection equations directly in the fragment shader to optimize rendering workloads.
Video Frame Interpolation & Upsampling (C++/Python)
Working on frame interpolation and resolution upsampling for custom videos, leveraging Nvidia DLSS (RTX 4090), optical flow, and NVOFA FRUC.
Meta Quest Simulation (Unity/C#)
Currently developing a VR simulation to practice and explore Meta Quest libraries within the Unity engine.
Past Projects:
Tree Predictor for binary classification (Python)
Implemented a tree predictor from scratch using Gini index/entropy, applied to classify poisonous mushrooms.
Ping Pong AI (C#)
Developed an AI agent using Unity's ML-Agents toolkit to learn Ping Pong via reinforcement learning.
Atrial Fibrillation Detector (Python)
Implemented a low-complexity algorithm for AF detection by applying signal preprocessing techniques to ECG data.
Point cloud acquisition using sensor (C++)
Implemented a custom C++ application to acquire and process point cloud data using the Occipital Structure Core sensor.
Technical Skills
AI & Computer Vision
- PyTorch, Pandas, Numpy, Sklearn
- Hugging Face Transformers & LMMs
- Distributed Training (torchrun)
- GPU Memory Optimization
- OpenCV, MediaPipe
3D & VR Development
- CUDA, Open3D
- Unity Engine
- VR Systems & Simulation
Web & Infrastructure
- FastAPI, Gradio
- Docker, AWS (Basic)
- PostgreSQL + PGVector
- Unix Shell, Vim
Languages
- Proficient: C, C#, Python
- Familiar: C++, Java, Bash