Founding Perception Engineer

We are hiring a Founding Perception Engineer

If you do not read this entire posting, do not apply.

This role is remote in the first phase, and we are open to international candidates. Where you are based matters far less than your relevance, technical depth, and ability to contribute to what we are building.

Important application requirement: Every application must include both a CV and a short written application. Applications without both will be rejected immediately. The written application should briefly explain who you are, what concrete relevance you have to Noctra, what you believe you can contribute, and why you are a serious fit for this role. Keep it short, clear, and specific.

Noctra is building a real-time AI system for situational understanding in environments where data is incomplete, latency is critical, and decisions have operational consequences. We already have a working MVP. This is not a research role, not a typical computer vision role, and not a place to learn fundamentals. It is for people who have already built perception systems that work, and now want to build one that actually matters.

The core problem is not detection. It is understanding what is happening over time, under noise, uncertainty, and degraded conditions. This means reliably tracking objects across time and occlusion, maintaining identity under imperfect data, handling false positives, false negatives, and inconsistent inputs, and extracting meaningful behavioral signals rather than just detections. Not in controlled environments, but in reality.

Today, the system ingests live video streams, performs detection and multi-object tracking, and maintains persistent object state. Where it breaks, and where you come in, is in unstable tracking under occlusion and noise, identity loss and re-identification failures, degraded performance in real-world conditions, limited behavioral understanding beyond detection, and sensitivity to imperfect and asynchronous inputs.

In this role, you will work directly on the perception layer of the system. Not isolated models or offline experiments, but a real-time system that must interpret the world continuously. Your work will include improving multi-object tracking robustness, building and refining re-identification logic, handling occlusion, noise, and degraded inputs, reducing false positives and false negatives, extracting higher-level signals from perception outputs, and making perception stable over time rather than just accurate per frame. You are responsible for enabling the system to understand continuity, not just detect objects.

We are looking for candidates who fall into one of two categories. Either you have relevant defence, ISR, or autonomy background from environments such as ISR or surveillance systems, defence platforms, autonomous vehicles or drones, or robotics perception systems; or you do not, but you compensate for it through having built perception systems that operate under real-time constraints, noisy and imperfect inputs, real-world deployment conditions, and environments where errors have consequences.

Technically, you likely have strong experience in several areas such as multi-object tracking, re-identification systems, computer vision for real-world environments, perception in robotics or autonomy, temporal modeling and sequence understanding, and model deployment in production systems. More importantly, you have dealt with tracking failures in practice, understand why perception systems break, think in terms of time rather than frames, optimize for stability rather than benchmark scores, and care about what happens after detection.

This is an early-stage, technically demanding role in an ambiguous environment, focused entirely on real-world performance. It is not a research position, not a clean dataset problem, not a benchmark optimization role, and not a place to work on isolated models. You will help build perception that works when conditions are not ideal.

We are currently pre-seed, with a raise planned from May to June. Compensation is equity-based at a founding level, and the role is remote initially with international candidates welcome.

What we are building matters because most systems stop at detection. We are building a system that maintains understanding over time, moving from detection to tracking to behavior to decision in real time.

We are not looking for many people, but for a small number of exceptional ones. If you are unsure whether you are a fit, you probably are not. If you know you are, reach out.

Please send both your CV and your short written application to: [email protected]