About
I'm a software engineer specializing in backend architecture and systems programming. I do my best work at the intersection of analytical theory and low-level execution - where rigorous problem-solving meets memory, state, and timing. I take pride in bringing order to chaotic environments through careful design of boundaries and coordination.
My engineering approach was shaped by two contrasting environments: two intensive years of CPGE math and physics, followed by the gritty, low-level systems architecture of 1337.
Today, my obsession with orchestration extends to leading teams. As Vice President of Praxis, I drive student teams to ship serious technical products. The ultimate test of this has been building MedicAtlas a full healthcare platform where I tackled the struggle of integrating DB architecture, dockerized services, and complex API layers into one reliable system.
You'll see this same systems-first mindset across my solo work, whether managing thread contention in Codexion or chaining AI decisions in Agency Brain. I'm currently seeking backend roles where scalable coordination is the core job. Outside of work, you can usually find me swimming, playing League of Legends, or immersed in a good book.
Experience
Systems Programming Student · 1337 / UM6P
Low-level C, concurrency, and systems, alongside applied AI - focused on orchestration and backend. The case studies below come out of this work.
- C
- Python
- POSIX threads
- PyTorch
- systems
CPGE MP (Maths & Physics) · Classe Préparatoire
Two years of intensive mathematics and physics - rigorous theory and the habit of sitting with a hard problem until it breaks.
- Mathematics
- Physics
- Algorithms
Projects
Agency Brain
A multilingual WhatsApp assistant for a study-abroad agency handling student questions in French, Arabic, and Darija. Replaced two hardcoded routing rules with constrained-decoding classifiers, pushing routing accuracy from 82% to 89% and keeping 50% of traffic away from the LLM entirely - cutting generation costs in half.
- RAG
- constrained decoding
- FastAPI
- ChromaDB
- Python
Codexion
A multithreaded C simulation where N independent threads compete for shared resources without ever deadlocking or starving. Engineered from scratch with POSIX threads, a hand-rolled min-heap scheduler, and a dedicated monitor that catches burnout within ~2ms - clean Valgrind exit, zero data races under ThreadSanitizer.
- C
- pthreads
- min-heap
- Valgrind
Fly-In
A collision-free drone traffic controller that routes 25 drones through a 54-hub, 70-edge map to a single landing zone. Built a custom space-time A* that solves the hardest map in ~55ms across 10 difficulty levels - zero collisions by construction.
- Space-time A*
- Dijkstra
- Python
- Pydantic v2
Projects related to the club
MedicAtlas
A full-stack patient platform (currently at MVP) where patients upload medical reports and get a structured, unified health history - from database architecture and dockerized services to a complete frontend-backend integration. AI extracts and normalizes data from scanned documents into trackable health timelines.
- FastAPI
- Next.js
- PostgreSQL
- Docker
- OCR
Arabic_Ocr
A CNN built from scratch in PyTorch to recognize all 28 handwritten Arabic letters. Two conv blocks, an 80/20 train/val split, and an Adam optimizer - the confusion matrix shows the model mostly trips on letters that look similar even to humans.
- PyTorch
- CNN
- torchvision
- scikit-learn