WORK · 2024 TO PRESENT
Two roles, both since 2024. The thread is ownership: writing software end-to-end and being on the call when something breaks.
ENGINEERING · SINCE 2024
Pixel38
Full Stack Developer
Beirut, Lebanon
JUL 2024 – PRESENT
The shipment platform was already running when I joined, and it kept running because someone made decisions every day. I became that person. One developer, one platform, one set of clients to talk to directly. Architecture calls, scope calls, the awkward email when a release slips: all on me. That posture changed how I wrote code. You ship differently when there is no one downstream to clean up after you.
Years of shipping under pressure had left the codebase dense with conditionals nobody could fully explain. I started at the worst file and worked outward, replacing tangled flows with smaller, named pieces, retiring dead branches, and pushing tests in front of the parts that mattered most. About forty percent of the technical debt came off the books over time. Core logistics workflows that nobody wanted to touch are now boring to read, which is the highest compliment a system like that can earn.
The trickiest piece was wiring the platform into enterprise accounting. Invoice sync had to be reliable across networks I did not control, schemas that moved without warning, and clients who expected the dashboard to keep responding while reconciliation happened in the background. I moved the integration onto a queue, took retries off the request path, and gave operators a way to inspect failures instead of guessing. The error rate fell.
Alongside that, I built the backend API and internal operations portal for a real-time food-delivery product: order intake, live driver tracking, restaurant onboarding, and the dispatcher screen that has to stay legible while a queue is moving. Today the work is mostly stewardship: keeping the platform stable, extending it where the business needs room, and resisting the urge to add anything it does not yet earn.
ENGINEERING · 2024
SlashML
Full Stack Developer
Remote
APR 2024 – JUL 2024
I spent three months at SlashML on a narrow brief: stand up an EC2 host for an LLM workload, on demand, without disturbing the rest of the fleet. LLM workloads are picky about hardware, model weights are large, and a static fleet either burns money on idle GPUs or runs out of capacity right when a customer needs one. Manual provisioning does not survive contact with real traffic.
I built the provisioner in Pulumi so the infrastructure was code that could be reasoned about and rolled back, then put it behind a queue so install jobs could be paced, retried, and inspected instead of fired blindly at AWS. New models came up cleanly; old ones got recycled without taking neighbors with them. To make the running system legible, I built a small internal dashboard for the team: which boxes were live, which models were on them, and where the deploys were stuck. Short engagement, but the kind of work where the constraints force you to think clearly about state, failure, and observability from day one.
If you are hiring, the email is below.