AI Systems · Security · Infrastructure
AI Systems Engineer
I build AI systems, break them on purpose to understand failure modes, and help decide which systems are safe to ship.
Accessible by design Keyboard Screen readers Reduced motion
Work and school, side by side.
Building and governing AI and automation systems in Enterprise Architecture, from production LLM pipelines to formal risk review of proposed AI deployments.
Graduate work across high-performance ML systems, systems engineering, and computer engineering.
Undergraduate foundation across software, computer systems, and computer engineering.
What I build with.
01
# ml-security
02
# systems-performance
03
# cloud-ai
Work / 02
ML · Security / 01
PGD reduced ResNet-34 top-1 accuracy from 76.0% to 0.2%.
Generated FGSM, PGD, and localized-patch adversarial examples against ResNet-34 on a 500-image ImageNet subset, then evaluated the same perturbed inputs on DenseNet-121. PGD used ε=0.02 for 10 iterations and produced a 90.50% cross-model degradation ratio.
ML Systems · GPU Performance / 02
MAX cut large-model GEMM latency from 3.49 ms to 1.83 ms.
High Performance Machine Learning project benchmarking PyTorch/CUDA, custom CUDA kernels, and Modular MAX on Transformer inference. MAX reached ~1.9× faster GEMM and ~1.12× faster end-to-end inference on the representative large model, while PyTorch remained faster on LogSoftmax and LayerNorm. My work included Mojo/MAX integration, experimental GEMM, Softmax, and LayerNorm kernel development, compatibility debugging, and benchmark analysis.
01 · Compute-bound
GEMM · large model
3.49 → 1.83 ms
MAX · ~1.9× faster
02 · Reduction-bound
LogSoftmax · LayerNorm
PyTorch
Faster
03 · End-to-end
Large-model inference
134.5 → 120.0 ms
MAX · ~1.12× faster
AI Systems / 03
Backlog → story draft → sync → retrieval.
Agents draft stories, sync work items, and retrieve answers from sprint documentation.
AI Systems · Review / 04
Document intake → rules → flagged review.
Extracts structure, checks explicit rules, and surfaces exceptions for human review.
AI · Governance / 05
First-pass risk check before human review.
Checks data sensitivity, third-party exposure, and access-control risk.
Field Archive / 03
Hackathons, professional communities, workshops, and the rooms where technical work becomes something shared.
Court Register
Record01of10
Browse with Previous/Next, arrow keys, trackpad, or swipe.
01 / Build
Hackathon build session.
02 / Industry
Learning beyond the day-to-day.
03 / Community
Engineering in company.
04 / Exchange
Technical conversations beyond the screen.
05 / Team
Building collaboratively under a deadline.
06 / Learning
Curiosity outside the boundaries of work.
07 / Community
The people behind the technical work.
08 / Record
A small artifact from the larger experience.
09 / Developer
Meeting the broader developer community.
10 / Learning
Looking beyond engineering alone.