Staff Applied Research Engineer, Biometrics

The Biometric Solutions team at Veriff develops the biometric and liveness technologies that sit at the critical entry point of our automated identity verification funnel.

Having established a robust baseline using existing SOTA architectures, we are now entering the next stage of our roadmap - institutionalizing a high-velocity "Fast-Follower" engine. This phase is focused on radicalizing our adaptation speed—moving from academic breakthroughs to production-grade, proprietary models in weeks, not months. Your role will bridge the gap between theoretical computer vision research and production code that protects millions of users.

You’ll help us enable Biometric innovation by:

• Architecting the Research Pipeline: Building the infrastructure to reproduce and adapt SOTA papers within 4–6 week cycles, ensuring our models outpace the global fraud landscape.
• Engineering Physics-Based Defenses: Developing "attack-agnostic" features—leveraging optical flow, specular reflection, and 3D geometry—to exploit immutable physical laws that attackers cannot easily circumvent.
• Solving for Domain Heterogeneity: Implementing advanced Domain Adaptation and Generalization techniques to eliminate performance variance across device types (iOS vs. Android), demographics, and environmental conditions.
• Optimizing for SaaS Scale: Writing high-performance, maintainable Python code to ensure complex CV models remain cost-efficient and latency-optimized.
• Strategic Prototyping: Working from first principles to decide when a novel architecture is required versus when the solution lies in superior data curation or procedural attack generation.

You are the right future Veriffian for the job if you have:

• 5+ years of experience in Applied Research or Data Science, with a proven track record of moving complex Computer Vision models from paper to production.
• Deep Mastery of CV Frameworks: Expert-level PyTorch or TensorFlow skills, specifically in implementing custom layers, loss functions, and domain-adversarial training loops.
• A "Physics-First" Intuition: Strong background in classical image processing (OpenCV) and 3D geometry; you understand why a 2D surface looks different than a 3D face to a sensor.
• Business & Production Acumen: You treat inference latency and compute cost as first-class constraints. You prefer an elegant, efficient solution over a "brute-force" ensemble.
• Strong Engineering Foundations: Proficient in Python, Bash, and SQL. You are comfortable building your own data pipelines and debugging containerized services.
• A skeptical, questioning approach: You don't take "SOTA" claims at face value. You investigate failure modes, hunt for bias, and demand empirical evidence before committing to a new direction.