About the Role
We are looking for an AI Research Engineer (Computer Vision) to research, prototype, and validate high-accuracy computer vision solutions on live CCTV footage.
This role sits at the intersection of research, experimentation, and real-world deployment. You will focus on building working prototypes in Jupyter notebooks, validating them on production-like data, and transferring knowledge to MLOps and Engineering teams for scaling and deployment.
This is a fast-paced startup environment with a strong emphasis on deliverables and outcomes, not fixed working hours.
Key Responsibilities
Research & Prototyping
• Research and prototype computer vision models and algorithms for real-world CCTV use cases
• Build and iterate on Jupyter notebook prototypes using real video streams and datasets
• Optimize models for high precision and recall under challenging conditions (occlusion, lighting, camera angles, motion blur, etc.)
• Experiment with detection, tracking, re-identification, temporal reasoning, and multi-camera setups
Validation & Metrics
• Design and run evaluation pipelines for model accuracy, precision, recall, and latency
• Analyze failure cases and propose systematic improvements
• Perform ablation studies and controlled experiments
Knowledge Transfer & Collaboration
• Conduct knowledge transfer (KT) sessions with:
• MLOps team to improve training pipelines, model performance, and inference efficiency
• Engineering team to translate research prototypes into deployable production systems
• Document research findings, assumptions, and trade-offs clearly
Deployment Awareness
• Collaborate with engineers to ensure research prototypes are deployable in production environments
• Consider real-world constraints such as compute limits, latency, GPU usage, and scalability
Required Qualifications
• Strong foundation in computer vision (object detection, tracking, segmentation, video analytics)
• Proficiency in Python
• Experience working with Jupyter notebooks for research and prototyping
• Solid understanding of precision, recall, F1, ROC, and evaluation metrics
• Ability to reason about real-world data noise and edge cases
• Comfortable working with unstructured video data
• Candidates must have a valid Singapore work permit or Singapore citizenship/PR.
Good to Have (Nice-to-Haves)
• Previous industry or research experience in computer vision
• Experience with NVIDIA DeepStream pipelines
• Familiarity with GPU inference optimization (TensorRT, CUDA concepts)
• Experience working with live CCTV or surveillance footage
• Exposure to MLOps workflows and model deployment pipelines
Work Environment & Culture
• Startup environment with high ownership and autonomy
• Deliverable-based working hours (not a strict 9–6)
• Fast iteration cycles and real-world impact
• Close collaboration with engineering and MLOps teams
• Opportunity to see research deployed into production quickly
Why Join Us
• Work on real production computer vision problems, not toy datasets
• See your research go live and create immediate impact
• High ownership, minimal bureaucracy
• Competitive pay with strong learning and growth opportunities
