Computer Science @ University of Minnesota
systems engineering · low-level programming · ML systems
I build efficient, dependable systems close to the hardware and useful in production.
ML Systems GPU Performance Computer Architecture
Distributed Systems Linux AI Infrastructure
+ ML systems and efficient AI on real hardware
+ GPU performance and computer architecture
+ Distributed systems and systems software| Project | What it is | Stack |
|---|---|---|
| TensorForge | Empirical GPU roofline modeling and performance analysis | Python · PyTorch · CUDA · NVML |
| QuorumKV | Raft-based distributed key-value store | Go · Raft · TCP |
| WireStack | TCP/IP networking stack built from scratch | C++ · Ethernet · IPv4 · TCP |
| Aegis | Secure GitOps private-cloud platform | Kubernetes · FluxCD · Cilium · Kyverno |
| CommerceCore | Transactional event-driven commerce backend | Java · PostgreSQL · Kafka · gRPC |
| Linux Kernel Lab | Kernel, QEMU, BusyBox, and device-driver experiments | C · Linux · QEMU · GDB |
languages:
- C++
- C
- Go
- Python
- Java
- TypeScript
- SQL
- Bash
systems:
- Linux
- TCP/IP
- Raft
- Distributed Systems
- Computer Architecture
infrastructure:
- Kubernetes
- Docker
- FluxCD
- Terraform
- GitHub Actions
data:
- PostgreSQL
- Kafka
- Redis
- Apache Iceberg
- Trino
ml-systems:
- PyTorch
- CUDA
- NVML
- MLflow
- Model ServingI like understanding what happens below the abstraction.
That usually means working somewhere between distributed systems,
operating systems, networking, GPU performance, and ML infrastructure.
I care about systems that are measurable, reproducible,
and actually work outside of a demo.
[engineering]
measure = "before optimizing"
automate = "what should be reproducible"
test = "the failure paths too"
document = "what future me will forget"
ship = "working systems over impressive diagrams"Portfolio · LinkedIn · TensorForge · QuorumKV · WireStack · Aegis
minh@github:~$ ▋