Mustapha amine D.

Production Systems Engineer: Firmware, Infrastructure & AI in Product

I build production systems that survive contact with the real world, not just the demo. Firmware, infrastructure, and the AI on top. The whole stack, not one slice of it. Most engineers pick an altitude and stay there: the app layer, or the model, or the hardware. My edge is that I don't. The hard problems always cross those lines, and I can hold the entire system in my head and make every layer work together. That range is the job. I stick around until it works, and after. If any of this sounds like you, we'll probably work well together. "We're bringing up new hardware and we need firmware from someone who protects the architecture, not just someone who makes it compile." Close to my best work. A client recently called my embedded work a masterclass. Not because the firmware ran, but because I caught the architectural risks before they reached manufacturing. "We keep scaling by throwing servers at the problem instead of finding the real bottleneck." Usually it isn't a capacity limit. It's one bad assumption. I've built fleet platforms running up to 25,000 machines and traced "we need more hardware" straight back to a single line of logic. "We shipped an AI feature that crushed the demo and fell apart in production." I build the system around the model, not just the model. Pipelines, guardrails, failure handling, the parts that decide whether it holds up under real traffic. I once pulled a legal-document AI from 44% to 95% accuracy in five days by fixing everything except the model. "Something is broken deep in the system and nobody can find it." This is what I'm best at. The bug is almost never where everyone's looking. I find the actual cause, not the convenient suspect. A few things that are true about working with me. You talk to me. Not an account manager, not a junior doing the real work while someone senior closed the sale. I'll tell you when your problem needs a smaller solution than you're picturing, even when the honest answer costs me the bigger invoice. Not everything needs AI, and I'll say so. I build the boring reliable version when that's right and the model-heavy version when it isn't. I've built for defense, medical devices, industrial automation, fintech, and early-stage startups. Different industries, one skill: Making things work where failure actually costs something. Recent work: -Embedded firmware architecture for a new hardware bring-up on Ambiq Apollo510B. The client called it a masterclass in embedded engineering. -A speech-to-command system for a tactical drone. 95.7% accuracy at 22ms, fully offline on edge hardware. -A server fleet monitoring platform managing up to 25,000 machines at once. -A legal-document AI rebuilt to know when it's actually unsure. Accuracy went from 44% to 95% in five days. -A factory-floor alarm system that was crying wolf. Cut false positives by 41%. Tell me what you're building or what's broken. You'll get a straight answer, not a sales pitch.