About Vikram R

About Me

Systems architect and AI researcher

I am a systems architect and AI researcher operating at the absolute lowest levels of compute. Currently, I am the solo founder and CEO of KarLex AI, Inc. (Delaware C-Corp), where I am building the next generation of hardware-native intelligence.

To make its continuous-control reinforcement learning fast enough, I research and design custom sub-quadratic neural architectures (like MoHM) using custom NVIDIA Triton and Apple Metal hardware kernels.

Current Core Focus

  • Sub-Quadratic Foundation Models: Solving exact-recall in Linear Attention and SSMs.
  • Compute Kernels: Writing bare-metal GPU optimizations in Apple Metal, MLX, and Triton/CUDA.
  • Operating Systems (Linux/eBPF): High-performance, zero-overhead kernel manipulation and networking.
  • Reinforcement Learning: Continuous control (TD-MPC2, CrossQ) for dynamic systems.

Key Focus Areas

Sub-Quadratic Foundation Models
Compute Kernels
Operating Systems (Linux/eBPF)
Reinforcement Learning