Software Engineer · ML Infrastructure · Backend Systems · AI Safety
triton-kernels — GPU kernels in Triton
- FlashAttention, MatMul, Softmax, LayerNorm
- Autotuning for T4/V100/A100/H100
aria — LLM Red-teaming Agent
- 10 attack strategies, 77 variants
- Reflexion-based learning from failures
ddp-lora-trainer — Multi-GPU LoRA Fine-tuning
- PyTorch DDP with gradient checkpointing
- Mixed precision training (FP16/BF16)
cuda-mempool — CUDA Memory Allocator
- Stream-ordered allocation
- Size-class binning
vibeclean-ai — Static Analysis for AI Code
- AST-based vulnerability detection
- OWASP Top 10 coverage
lockfree-queue — Lock-free Queues (Rust)
arena-alloc — Memory Allocators (C++)
Languages: Python, C++, Rust, SQL ML/AI: PyTorch, Triton, CUDA Infrastructure: Docker, Kubernetes, AWS Data: Kafka, PostgreSQL, Redis