Lead Computer Vision Engineer – Real-Time Video & Action Recognition

You lead the computer vision work behind a large-scale real-time video analytics and quality assurance platform. You design the models, take them to production and set the technical direction for a growing team. 

About K&C

K&C (Krusche & Company) is a Munich-based technology partner for medium and large European companies. Our senior engineering teams in Poland, Ukraine, Bulgaria and Egypt build, modernize and scale business-critical software for clients such as Nestlé, Bosch, Sparkassen and Informa. 

Our promise is Beyond Outsourcing®: deep technical ownership, German business standards and engineering excellence. K&C AI Solutions brings this to computer vision, machine learning and AI transformation. 

The project

A global live-video platform operates dozens of concurrent video streams around the clock. Each session follows precisely defined operational procedures involving object positioning, hand movements, interaction sequences and workflow compliance. 

You will build the computer vision system that monitors these procedures in real time and transforms observations into actionable quality and process insights. 

• 50 concurrent video streams with sub-second detection 

• Around 40 procedure categories, from object-position verification to fine-grained hand-motion analysis 

• Three core models combined with a rule engine and event-driven processing 

• Runs on AWS inside the client’s cloud environment 

• Initial scope focuses on object-state verification and positioning checks, followed by hand tracking, gesture recognition and advanced action understanding 

What you will do

• Own the end-to-end vision architecture: object detection, hand-keypoint tracking, temporal action recognition and the rule engine that combines them 

• Select, train and fine-tune models such as RT-DETR, RTMPose and skeleton-based action recognition on large-scale production video data 

• Optimize inference for real-time multi-stream production with TensorRT, Triton and DeepStream or comparable tools 

• Define the annotation strategy: label definitions, formats, volume per class and staged recordings for rare events 

• Set precision and recall targets per procedure category and build evaluation frameworks stakeholders trust 

• Build the feedback loop that turns operational decisions into new training data 

• Translate operational workflows into measurable detection logic together with product, operations and engineering teams 

• Mentor ML engineers and shape the team as it grows 

What you bring

• 6+ years in computer vision, including 3+ years taking video models to production 

• Production experience with human pose, hand or gesture recognition in video 

• Hands-on work with temporal action recognition or activity detection 

• Real-time inference at scale: multiple concurrent streams, tight latency budgets and GPU optimization 

• Strong Python and C++ for performance-critical paths; PyTorch, ONNX and TensorRT 

• A proven approach to rare-event detection and imbalanced classes 

• Cloud deployment experience, ideally AWS (EKS, SageMaker, S3) 

• Fluent English and the ability to explain model capabilities clearly to product and business stakeholders 

Strong plus

Experience from fields that solve closely related challenges gives you a head start: 

• Industrial assembly verification or worker-step compliance 

• Sign language recognition 

• Surgical video analysis with hand and instrument tracking 

• Retail self-checkout and hand-object interaction 

• Sports analytics with high-speed pose tracking 

• Manufacturing quality assurance using computer vision 

What K&C offers

• A flagship real-time AI project with visible impact on a global platform 

• Architecture authority and full technical ownership of the vision stack 

• A long-term engagement with a clear path to additional use cases, environments and team growth 

• Senior peers across K&C’s European engineering hubs 

• Competitive compensation that reflects senior expertise 

How we hire

1.      Intro call with K&C, 30 minutes

2.      Technical deep dive, 60–90 minutes, on real problems from the project: occluded objects, rare events, multi-stream throughput

3.      Meet the client team

4.      Decision within one week

To Apply 

Send your CV and a short note describing one production computer vision system you built: the problem, the scale and the outcome. 

Apply via [link or email].