Computer Science Graduate · MSc Data Science Student · Software Developer
Bucharest, Romania
I build software at the intersection of scientific computing, machine learning, computer vision, and real-time 3D visualization. I enjoy turning technically challenging ideas into working applications, with an emphasis on measurable results, clear engineering, and open-source development.
An open-source, interactive 3D simulator for solar and lunar eclipses, built with Unreal Engine 5.7, Cesium for Unreal, C++ and a Python/Skyfield backend powered by JPL DE441.
- Catalogue of 71,264 solar and 72,272 lunar eclipses, spanning approximately 30,000 years (13,000 BC–17,000 AD).
- Renders 3D umbra and penumbra cones and their interaction with a WGS84 ellipsoid, with geospatial terrain from Cesium.
- Computes solar-eclipse contact times (C1–C4), shadow paths, and city visibility information.
- Includes a lunar-limb elevation model trained on NASA LRO/LOLA data; reported validation RMSE against the gridded DEM: 12.57 m.
- Contact-time comparisons against Xavier Jubier for three selected total solar eclipses showed a maximum difference of 1.7 seconds in the documented test cases.
Source code · Trailer · Full demo · Download · itch.io
♟️ AI Chess Bot
An AlphaZero-style chess engine combining a residual policy-and-value network with Monte Carlo Tree Search (PUCT).
- Trains a PyTorch network on Stockfish-evaluated positions, then exports it to ONNX.
- Runs in the browser using
onnxruntime-web, with WebGPU and WASM fallback; inference runs in a Web Worker. - Includes a desktop application for training, match arenas, model comparisons, and strength testing against Stockfish.
- Documents benchmark results and their limitations, including the uncertainty of Elo estimates from the current match sample.
An implementation and performance-study project exploring the Canny edge-detection algorithm, from a serial baseline to parallel variants.
- Implementations using CUDA, OpenMP, Pthreads, MPI, and hybrid MPI + OpenMP.
- Covers the main Canny stages: Gaussian smoothing, Sobel gradients, non-maximum suppression, double thresholding, and hysteresis edge tracking.
- Includes profiling scripts and comparative performance analysis across implementations.
Languages
C++ · Python · C, Java, Matlab, .NET · HTML/CSS, Assembly
Machine Learning & AI
PyTorch · ONNX · Neural Networks · Monte Carlo Tree Search
Scientific & Geospatial Computing
Skyfield · JPL DE441 · Cesium for Unreal · WGS84 · Numerical Geometry
Parallel & High-Performance Computing
OpenMP · MPI · Pthreads · Profiling
3D & Interactive Applications
Unreal Engine 5 · Real-Time Rendering · Scientific Visualization
- MSc Data Science — National University of Science and Technology POLITEHNICA Bucharest (currently pursuing)
- BSc Computer Science — National University of Science and Technology POLITEHNICA Bucharest
- Honorable Mention — Student scientific session, May 2026, for the eclipse-simulation project.
- GitHub: @Dragos112358
- LinkedIn: Dragoș Boncan
- Website: dragos112358.github.io
Interested in scientific software, computational astronomy, machine learning, parallel computing, and real-time visualization. Open to constructive feedback, technical discussions, and collaboration.