░█░█░█▀▀░█░█░░░▀█▀░█░█░█▀▀░█▀▄░█▀▀
░█▀█░█▀▀░░█░░░░░█░░█▀█░█▀▀░█▀▄░█▀▀
░▀░▀░▀▀▀░░▀░░░░░▀░░▀░▀░▀▀▀░▀░▀░▀▀▀
Phani Kumar
Software Engineer | ML Researcher | Systems Builder
👨💻 About Me
Software engineer with expertise in machine learning, full-stack development, and systems programming. Passionate about building high-performance systems that bridge classical algorithms with modern ML techniques.
📍 Location: Vijayawada, India
🎯 Status: Open to full-time opportunities
💼 Focus Areas: ML Engineering • Full-Stack Development • Systems Programming
Professional Mission: Developing innovative solutions at the intersection of machine learning, computer graphics, and high-performance computing. Seeking opportunities to apply these skills in production environments — whether building ML infrastructure at scale, designing full-stack applications that handle millions of users, or optimizing low-level systems for performance. Particularly interested in teams that value versatility, technical excellence, and end-to-end ownership.
🎓 Highlights
✨ Featured in Hitesh Malhotra's AI Challenge 2025 (Optimizing-Ray-Tracing project)
🚀 4+ Production-Ready Projects spanning ML, systems, and full-stack development
⚡ Specialized Expertise in ML-guided rendering, GPU acceleration, and compiler design
📊 Research Focus on hybrid classical-ML approaches for computational efficiency
🛠️ Technology Stack
Core Technologies:
- ML/AI: PyTorch, TensorFlow, GANs, Diffusion Models, Transformers
- Systems: Rust, C/C++, CUDA, GPU Programming
- Full-Stack: Go, Svelte, React, Node.js, PostgreSQL, Supabase
- Tools: Docker, Git, Linux, Blender, Mitsuba 3
Specializations: Machine Learning for Graphics • High-Performance Computing • Compiler Design • Distributed Systems
🚀 Featured Projects
| Project | Description | Technical Highlights | Impact |
|---|---|---|---|
| Optimizing-Ray-Tracing | ML-guided path tracing renderer combining classical ray tracing with learned path generation and diffusion-based refinement. Featured in AI Challenge 2025. | PyTorch, CUDA, GANs, Diffusion Models, Physics-based Rendering | Achieved significant variance reduction in rendering convergence through hybrid ML-classical approach |
| ASTRA | Astronomical Spectral Transformer for galaxy redshift approximation using custom attention mechanisms. | PyTorch, Transformers, RoPE Embeddings, Mixture of Experts (MoE) | Developed novel architecture for high-accuracy spectral analysis in astrophysics domain |
| AstroPixel_Spectra | Curated dataset of 3,500+ galaxies with multi-filter images and corresponding spectra for ML research. | Data Engineering, Scientific Computing, SciPy, NumPy | Enables multi-modal ML research in astronomical data analysis |
| c2bf-in-RUST | C-to-Brainfuck compiler demonstrating end-to-end compiler design with custom parser and IR. | Rust, Compiler Design, Pratt Parsing, Intermediate Representation | Showcases deep understanding of language design and compilation pipelines |
| Caveman Board | Real-time multiplayer physics simulation engine built from scratch with adaptive spatial optimization. | Go, WebSockets, Spatial Hashing, Concurrency | Demonstrates systems programming and real-time distributed architecture skills |
💡 Technical Approach
Philosophy: Building robust, scalable systems that combine theoretic