icetana is a global leader in AI-powered video analytics, delivering real-time anomaly detection solutions to enhance situational awareness and security for critical infrastructure, campuses, smart cities, and large enterprises. Our technology helps clients monitor vast camera networks with AI that highlights abnormal behaviors, dramatically reducing the burden on human operators.
If you’re excited about writing software that is close to the metal, where performance matters, code that is reliable, fault tolerant and observable, this is the company for you.
If you think memory alignment is beautiful and undefined behavior is terrifying, let’s talk.
Your mission is to design and build robust, high-performance systems that power our computer vision pipelines; enabling real-time, scalable, and efficient processing on the edge and in the cloud. You’ll work with C++, CUDA, and other low-level systems to squeeze every last drop of performance from the hardware. You care deeply about memory, latency, and throughput and have strong opinions about zero-copy, cache coherence, and lock-free data structures. You’re not just comfortable working close to the metal: you prefer it.
A big part of your role will involve making complex systems observable and debuggable. You'll add metrics, traces, and logs that help diagnose performance bottlenecks and correctness issues. You'll design your code to be testable, monitorable, and safe to evolve.
This is a small company that moves fast, you’ll likely have to wear many hats, there’ll always be opportunities to learn and grow and have a genuine impact on the team as well as the product we ship.
We have a fairly modern and light-weight software development process. Our tech stack has a number of services deployed in a single docker compose script including C++/Gstreamer for video processing, Py/Pytorch/YOLO/Ultralytics on the AI side, React/TS frontend, Python backend, PG database all running on rack mounted servers with 4x RTX 5000 GPUs in each.
We built a new stack for Facial Recognition and License Plate Recognition in the space of 6 weeks. Our team of only a handful of developers moves fast despite the challenging variety of problems we face when building ML heavy applications.
.webp){kind=logo}