Python · Django · FastAPI · Distributed Systems
Six years building and scaling production systems in Python — APIs, data models, and the infrastructure under them. Still learning something new most weeks.
I'm a backend engineer drawn to the unglamorous part of software — the part that has to not fall over when traffic spikes. I spend most of my time on APIs, data modeling for high-throughput workloads, and systems end-to-end, from the schema to the deploy.
Most recently I've worked on backend and architecture at kotook.ae, helping scale and maintain a PropTech platform and building its internal CRM. Before that I rebuilt a high-traffic store on Django and grew from developer to technical lead over four years.
Lately I've been wiring AI into backends — RAG search, vector databases, decoupled model inference — including my open-source project VectorLens. I'm still figuring a lot of it out, but I like systems that are clean, observable, and boring in the best way.
Own backend architecture and technical direction across the web platform, internal CRM, and cloud infrastructure. Built and scaled the backend, REST APIs, and data models from scratch; sustained sub-350ms latency at 1M+ requests with full observability. Shipped a RAG property-search assistant into core services.
Re-architected a legacy WordPress store into a scalable Django platform for high-traffic sales campaigns. Delivered a highly available, distributed system handling tens of millions of monthly requests at peak — improving performance ~50%, cutting infra costs 60%, reducing storage 75%.
Grew from Django developer to technical lead over ~4 years, building products end-to-end on Django/DRF — system design, data modeling, and CI/CD across multiple services. Represented the company at Dubai GITEX 2023.
An end-to-end FastAPI backend for text/image search using SigLIP 2 embeddings over PostgreSQL + pgvector. The interesting part is the architecture: API traffic is decoupled from heavy GPU model inference via Apache Kafka + Celery, with Server-Sent Events streaming live progress. Adds open-vocabulary object grounding (Grounding DINO / LLMDet), secured with JWT + RBAC and Docker multi-stage builds.
A PropTech platform I helped architect and scale from the ground up — a property marketplace plus the internal CRM that runs the sales team — with a RAG-based property-search assistant built on top.
A high-traffic e-commerce platform rebuilt for scale — horizontal scaling, disaster recovery, and full observability, load-tested with Locust to tens of millions of monthly requests.
A two-person team building an end-to-end ML pipeline to detect and classify failures in industrial cooling equipment from vibration, sound, and temperature sensor data. Our model came out with the top accuracy score in the competition, and we finished 2nd overall — alongside strong teams from Sharif University of Technology and other institutions. I learned a lot from the people we were up against.
The ACID tax, Redis's single-threaded design, B-Trees vs LSM Trees, compaction and Bloom filters — fix a bottleneck instead of moving it.
2026-06-19Two design choices do most of the work — Sequential I/O and the Zero-Copy principle. How each turns a disk into a firehose, with diagrams.