Hussein ChediOpen to new roles · Remote or Dallas, TX~/hussein-chedi — AI/ML Solutions Architect
I'm Hussein Chedi — I design distributed training and LLM inference on GPU cloud, most recently for IBM's enterprise customers.
p95 latency
−0%
vs. baseline
GPU utilization
0%
from 58%
Training time
−0%
vs. baseline
$torchrun --nproc-per-node 8 train.py --strategy fsdp
nccl all-reduce · ring · 8 ranks ok
gpu util 58% → 87% ✓ healthy
Shipped at
0+
Years of experience
0
Enterprise workloads
$0K
GPU spend saved / yr
I design the systems between a model and the business — training pipelines, low-latency inference and the platforms that run them.
Since 2016
0
years building ML systems & backend software
Dallas, TX
Remote across US & LATAM time zones
B.Sc. Computer Science
The University of Texas at Dallas, Richardson, TX | Aug 2015 - May 2019
Career path
2016
Software Engineer
Intellectsoft
2020
ML Engineer
Devsu
2022
Solutions Architect
IBM
04
live products
AI/ML & full-stack engineering
AI-designed interiors that customers can shop.
My part
I built the ML pipeline behind the room designer: models fine-tuned in PyTorch and Hugging Face read a prompt, photo or sketch, then match every piece to the retailer's real SKU catalog. I served them as GPU inference on Kubernetes, so a styled, shoppable room comes back while the customer is still browsing.
One-click AI virtual staging and interior design.
My part
I worked on the generative imaging engine behind staging and declutter — training image models in PyTorch on CUDA GPUs, then packaging them as containerized inference services. The same services power the one-click web app and the developer API, and scale on Kubernetes as listing volume grows.
AI stories where your child is the hero.
My part
I designed the story pipeline: an LLM writes each personalized chapter, then narration and illustration models turn it into a book a child can read and hear. I served the LLM with vLLM to keep stories quick to generate, and built the Python backend the iOS and Android apps call.
AI automation that changes how a business runs.
My part
I architected the platform the agents run on — LLM serving with vLLM on Kubernetes, connected to each client's tools and workflows. MLOps pipelines in MLflow and GitHub Actions let every new chat or voice assistant be tested and shipped safely, then run around the clock.
Each role a step closer to the hardware and the customer.
Designed GPU training and inference platforms for enterprise customers on hybrid cloud.
−35%
training time
−42%
p95 latency
$180K
yearly GPU savings
87%
GPU utilization
Stack
From CUDA up to Kubernetes — chosen for what it does in production.
Hiring for ML platform or inference work — or want a second look at a GPU bill? My inbox is open.
Email me directly
I can help with
