"Stay Hungry Stay Foolish." β Steve Jobs
I'm an AI/ML Solutions Architect with 10 years of experience taking machine learning from notebook to production. I design GPU training and inference platforms that train faster and serve for less β most recently at IBM, for enterprise customers on hybrid cloud. I love the moment a slow, expensive model turns into a fast, affordable one, and I enjoy teaching teams how to get there.
| β±οΈ β35% training time | Distributed training with PyTorch FSDP + Kubernetes across 12 enterprise workloads |
| π β42% p95 latency | LLM inference tuned with vLLM + Triton on NVIDIA GPUs |
| π° $180K / year saved | Annual GPU spend cut for customers through inference optimization |
| π 58% β 87% GPU utilization | NCCL and CUDA memory fixes in multi-node LLM training |
| ποΈ 3 days β 4 hours | Environment setup with Terraform + Helm for GPU node pools and Slurm |
| π 2M daily predictions | Containerized ML services with Docker + Kubernetes, deploys 60% faster |
|
Python |
Go |
C++ |
Java |
PyTorch |
TensorFlow |
scikit |
|
Docker |
Kubernetes |
Terraform |
AWS |
Actions |
Git |
GitHub |
| Product | What it does |
|---|---|
| Zory AI | AI-designed interiors that customers can shop, built from real retail catalogs |
| Collov AI | One-click AI virtual staging and room declutter for real estate |
| DreamKid | AI-written, narrated and illustrated stories where your child is the hero |
| DevoteAI | AI agents, chat and voice assistants that automate business workflows |
Hiring for ML platform or inference work, or want a second look at a GPU bill? My inbox is open.