SOPilot develops AI that turns videos of manual industrial processes into accurate, structured and multilingual Standard Operating Procedures (SOPs). Our goal is to make industrial knowledge easier to capture, understand and transfer. We combine computer vision, vision-language models (VLMs) and large language models (LLMs) to understand human activities and generate reliable work instructions. We are an early-stage startup at the TU Berlin Incubator, looking for an AI Research Engineer to join our founding team and shape the core technology behind SOPilot.
Your role
You work at the intersection of applied AI research and engineering, on hard problems in video understanding, computer vision and multimodal reasoning. You research, experiment with and improve state-of-the-art AI approaches to generate precise and reliable SOPs from real-world videos, and you turn promising research into practical solutions.
What you'll work on
- AI approaches for video understanding, action recognition and workflow segmentation
- Computer vision models, VLMs and LLMs for object detection, activity understanding and automated SOP generation
- Evaluating and optimizing AI models for accuracy, efficiency and reliability
- Experiments and benchmarks on real workstation videos
- Deploying AI solutions on cloud or local infrastructure
What we're looking for
- A strong foundation in machine learning and computer vision (object detection, segmentation, tracking, action recognition)
- Experience with VLMs, LLMs or multimodal AI
- Good Python skills and familiarity with PyTorch
- Cloud infrastructure (AWS, GCP, Azure or VPS) and Docker
- Curiosity, independent problem solving and enthusiasm for applied AI research
What we offer
- Equity and the opportunity to become part of the founding team
- Participation through the Berliner Startup Stipendium, subject to eligibility and funding confirmation
- Workspace, mentoring and access to the TU Berlin Science & Startups ecosystem
- Challenging research problems taken from real industrial tasks
#J-18808-Ljbffr