We are looking for a technically robust Deep Learning Engineer with hands-on experience in Computer Vision and Vision-Language Models (VLMs) to build and deploy production-grade video analytics solutions.
The role involves owning computer vision solutions end-to-end — from model development and training to inference optimisation and deployment on live, multi-camera environments. The position requires strong Python and C++ skills and experience working with real-world camera streams.
Key Responsibilities
- Build, train and evaluate computer vision models for detection, tracking, classification and segmentation.
- Design and deploy end-to-end video analytics pipelines for live camera streams.
- Develop and optimise multi-camera tracking across overlapping and non-overlapping camera views.
- Apply Vision Language Models (VLMs) for real-world scene understanding, including prompting, grounding, evaluation and pipeline integration.
- Optimise deep learning models for real-time inference on GPU and CPU infrastructure.
- Build and automate data pipelines covering ingestion, annotation, augmentation and dataset versioning.
- Debug and improve model performance on challenging real-world footage, including poor lighting, occlusion, camera drift and varying camera angles.
- Deploy,
monitor and continuously retrain models in production.
Requirements
Key Requirements
- Strong hands-on experience in Deep Learning and Computer Vision.
- Proven experience deploying computer vision models in production environments.
- Experience working with live camera/video streams.
- Strong programming skills in Python and C++.
- Experience with object detection, tracking, classification and segmentation.
- Experience building end-to-end video analytics pipelines.
- Hands-on exposure to multi-camera tracking.
- Experience with Vision-Language Models (VLMs) and practical applications of prompting, grounding and evaluation.
- Experience optimising models for real-time inference on GPU/CPU.
- Strong understanding of model training, evaluation, deployment and monitoring.
- Ability to troubleshoot models against real-world video quality challenges.