17-year-old AI Researcher focused on designing and training language model architectures from scratch. I build systems to understand the why behind learning dynamics — not just replicate them.
- Current Focus: Collaborative Expert Systems (MoC) for emergent reasoning in sparse models
- Specialties: Deep learning optimization, distributed training, experimental architectures
- Goal: Join a team that values technical rigor, creativity, and ambitious R&D
Languages & Frameworks: Python, PyTorch, Custom Autograd Functions
Architectures: Mixture-of-Experts (MoE), Sparse Attention (NSA), Grouped-Query Attention (GQA), QK-Norm, RoPE
Training at Scale: DDP, FSDP
Tooling: Weights & Biases, Docker, Git
| Lunaris Codex | Lunaris Codex MoC |
|---|---|
| Modular LLM training toolkit featuring SOTA Dense models and Hybrid NSA-MoE architectures. | Novel "Mixture-of-Collaborative-Experts" designed for emergent reasoning via 2-Pass communication. |