A data-oriented C++23 aerospace simulation engine and GNC framework for 6-DOF missile engagements, featuring Vulkan GPU compute and a live terminal telemetry cockpit.
Yashwanth
Studying Computer Science
I'm interested in software engineering, computer systems, and machine learning.
Projects
A Physical Optics radar cross-section (RCS) database generator in C++20/CUDA, computing high-frequency stealth signatures from 3D meshes for StrikeEngine.
A C++20 geospatial terrain compression library and CLI featuring adaptive predictive coding, quadtree partitioning, and random-access Region-of-Interest decoding.
A lean inference engine and CLI running the Qwen3.6-35B MoE model on consumer 8 GB GPUs via tiered mmap streaming, C++ AVX2 expert execution, and GPU LRU caching.
Writing
-
Treating LLM inference as a systems memory hierarchy problem: tiered mmap streaming, C++ AVX2 SIMD kernels, GPU LRU caching, and hybrid attention.
-
Designing a high-throughput C++23 flight simulation and GNC engine using Structure-of-Arrays memory, 1 ms physical truth, and Vulkan GPU compute acceleration.
-
How StrikeCEM generates monostatic radar cross-section databases from 3D surface meshes using Physical Optics, PTD fringe correction, and CUDA acceleration.
-
Designing the .xtm geospatial format: combining rate-distortion quadtree partitioning, 6 deterministic predictors, and sub-millisecond Region-of-Interest decoding in C++20.
Contact
- Email: yashwk@proton.me
- GitHub: github.com/yashwk