Senior Computer Vision Engineer

Location: Berlin (Onsite)

About Us

We are a defense technology startup based in Berlin, building the next generation defense layer against aerial threats. Our team develops mission-critical systems that combine hardware, software, sensing, and real-time AI to operate reliably in complex real-world environments.

Operating in stealth mode, we place strong emphasis on engineering rigor, real-world performance, and defense-grade system reliability.

Role Overview

We are seeking a Senior Computer Vision Engineer to design, train, deploy, and continuously improve computer vision models for aerial threat detection, classification, and tracking.

This role focuses on real-world vision challenges: detecting small objects at long range, maintaining tracking stability under poor visibility and motion, classifying targets reliably, and deploying performant models on edge compute platforms such as NVIDIA Jetson / Tegra.

You will own key parts of the vision model lifecycle, from dataset preparation and augmentation to model development, deployment optimization, field-data error analysis, and iterative improvement. You will work closely with robotics, tracking, and controls engineers to ensure visual perception outputs can be used reliably in real-time system behavior.

This is a hands‑on, production‑oriented role for someone with deep experience building vision systems that work outside controlled lab conditions.

This position requires onsite work in Berlin and candidates must be from a NATO member state.

Key Responsibilities

• Design, train, fine‑tune, and deploy computer vision models for: Long‑range small‑object detection, Object classification and identification, Multi‑object tracking, Target state estimation support
• Work with modern vision approaches, including real‑time detectors, CNN‑based architectures, vision transformers, tracking‑by‑detection pipelines, or custom models
• Evaluate trade‑offs between detection accuracy, latency, robustness, and compute constraints
• Develop model pipelines that support continuous improvement from real‑world field data

Real‑World Robustness

• Improve model performance under challenging operating conditions, including: Small targets at long range, Motion blur, Low light, glare, and changing illumination, Clouds, rain, fog, and atmospheric effects, Cluttered backgrounds, Camera movement and vibration, IR / thermal imagery artifacts
• Analyze failure cases and translate them into concrete model, data, and preprocessing improvements
• Improve classification confidence, tracking stability, and robustness against false positives

Dataset Lifecycle & Synthetic Data

• Own key parts of the vision data pipeline, including: Data cleaning, Augmentation strategies, Label quality analysis, Error analysis, Continuous dataset improvement
• Use synthetic data to improve coverage of rare, difficult, or safety‑critical scenarios
• Collaborate with simulation engineers to align synthetic data generation with real‑world failure cases
• Define metrics and evaluation sets for long‑range detection, classification, and tracking performance

Tracking & Sensor Fusion

• Develop and improve tracking‑by‑detection and multi‑object tracking pipelines
• Integrate vision outputs into downstream tracking and control systems
• Work with sensor inputs including: IR / thermal cameras, Laser range finder data
• Contribute to sensor fusion approaches where visual detections must be combined with additional measurement sources
• Collaborate with robotics, tracking, and controls engineers to ensure perception outputs are stable, timely, and actionable

Edge Deployment & Optimization

• Deploy and optimize models on NVIDIA Jetson / Tegra platforms
• Optimize inference performance using: TensorRT, CUDA
• Balance accuracy, latency, memory usage, and thermal constraints
• Support real‑time perception pipelines running on embedded or edge compute hardware

Required Qualifications

• 6+ years of experience in computer vision, applied machine learning, or real‑world perception systems
• Proven hands‑on experience building, training, deploying, and maintaining computer vision models
• Strong experience with object detection, classification, and/or multi‑object tracking
• Experience with real‑world image data, including noisy labels, domain shift, sensor artifacts, and difficult edge cases
• Strong Python skills and experience with deep learning frameworks such as PyTorch or TensorFlow
• Experience optimizing and deploying models on edge platforms, ideally NVIDIA Jetson / Tegra
• Experience with inference optimization tools such as TensorRT and/or CUDA
• Strong analytical skills for error analysis, model evaluation, and performance debugging
• Ability to work closely with robotics, controls, and systems engineers
• Onsite availability in Berlin
• Citizenship of a NATO member state

Preferred Qualifications

• Background in defense, aerospace, robotics, autonomous systems, or industrial perception
• Experience with IR / thermal computer vision
• Experience with long‑range small‑object detection
• Experience with sensor fusion involving visual detections and range measurements
• Experience with synthetic data generation and simulation‑driven dataset improvement
• Experience with tracking filters, tracking‑by‑detection, or real‑time perception‑to‑control pipelines
• Strong C++ skills for production integration are a plus

What We Offer

• VSOP / equity participation
• Central Berlin office
• Extreme growth opportunity in a fast‑scaling startup
• Monthly voucher
• Public transportation subsidies

Who You Are

• Real‑world focused: You care about models that work in the field, not just on benchmark datasets
• Data‑driven: You analyze failures systematically and improve models through evidence
• Performance‑aware: You understand latency, memory, and edge deployment constraints
• Robustness‑oriented: You think deeply about edge cases, domain shift, and degraded conditions
• Collaborative: You work effectively with robotics, tracking, and controls engineers
• Mission‑aligned: You understand the responsibility of building perception systems for defense applications