Software Engineer · ML Systems & Efficient Architectures · Student
I'm interested in the intersection of model architecture and inference efficiency — specifically, why modern sequence models are designed the way they are, where they fail, and whether simpler alternatives can close the gap.
Right now I'm building PULSE — an experimental O(n) architecture that replaces the standard SSM + Attention + State stack with a single uniform block. It's not finished. It's a research project where I'm learning by building from scratch.
Current focus areas:
- Linear attention and its approximation trade-offs
- Kernel-based sequence models vs. softmax attention
- Hardware-aware algorithm design (memory hierarchy, tiling, compute efficiency)
- Why Flash Attention works at the kernel level — working through the math
PyTorch Python Research
An experimental sequence architecture exploring whether a single O(n) primitive (local convolution + linear attention + gated fusion) can replace the complexity of transformer-style stacks. Active development — see the repo for current status and known issues.
C++17 P2P Distributed Systems · Archived
Distributed peer-to-peer file sync with ML-based anomaly detection, delta-sync algorithms, and self-healing network topology. No longer maintained — kept for reference.
| ML / Research |
|
| Languages |
|
| Web |
|
| Infra |
|
