Roman

I study CS, Machine Learning, and Math at Carnegie Mellon. I like problems that bring together math, algorithms, and the physical world. Lately that's meant perception: cameras, sensors, and motion.

I. Work

incoming · spring 2027

Tesla AIAI Engineering Intern

Joining the geometric-vision team to work on perception for Autopilot and Optimus.

may – aug 2026

OnDeskFull Stack Engineering Intern

Shipped two production agents that turn emails and meeting transcripts into structured project updates, with human review before publication. Also built a screenplay comparison tool that cut review time from hours to minutes.

may – oct 2025

Fuzzy Door TechSoftware Engineering Intern

Built an OpenCV camera-calibration tool used on Disney, Paramount, and Universal productions, then led calibration on set in England. Implemented real-time optical flow for live film feeds and researched camera–IMU synchronization; the work resulted in a U.S. patent.

u.s. patent no. 12,597,167 · issued april 2026

II. Projects

Optimizing compilerrust

Compiles a C-like language to x86-64 through an SSA-based IR, graph-coloring register allocation, and twenty-odd optimization passes. Outperforms GCC -O1 on most of the course benchmarks.

Neural network from scratchc++

Built a multilayer perceptron and matrix library in C++ using only the standard library, including forward propagation, backpropagation, and SGD.

Dynamic memory allocatorc

64-bit malloc, free, and realloc built on a footerless segregated free list with miniblocks.

Reinforcement learning agentspython · pytorch

Implemented Snake and 2048 environments in Pygame and trained PyTorch agents to play them.