Senior CV/ML Engineer
End-to-End Computer Vision Systems
The Opportunity
We are building a Computer Vision platform that automates the full CV model lifecycle, from data preparation and model training through evaluation, deployment, and monitoring.
You will join a small, autonomous team of 3 to 5 engineers (plus interns) working alongside AI/ML engineers and computer vision scientists at ST Engineering.
You are the subject matter expert for Computer Vision across the entire lifecycle: data preparation, model training, evaluation, deployment, monitoring, and automation. You will also build CV proof-of-concepts to validate product ideas before we commit to building at scale.
Scope: 2D image and video. This role does not cover 3D vision, point clouds, LiDAR, or sensor fusion.
What You'll Build
Computer Vision Models and Pipelines (Primary Focus)
- Train, fine-tune, and evaluate CV models for classification, detection, and segmentation on real, imperfect data
- Own the data side: dataset curation, labeling strategy and quality control, augmentation pipelines, and dataset versioning
- Define evaluation methodology that reflects the use case (mAP, IoU, precision and recall at the operating point), run error analysis, and build regression tests that stop weak models from shipping
- Package and deploy models to production (ONNX Runtime, TorchServe, or similar), including inference optimization against latency and throughput targets
- Monitor deployed models: performance and data drift, failure case capture, and retraining triggers
Automation Platform
- Build backend services (Python, FastAPI) that turn the above into repeatable, automated pipelines the team and the platform can trigger programmatically
- Wrap CV toolkits (PyTorch, OpenCV, Ultralytics YOLO) so data prep, training, evaluation, and deployment can be orchestrated end to end
- Design clean, versioned, documented APIs that internal teams and platform services consume
- Implement orchestration for multi-step pipelines with retry, rollback, and defined SLAs
Proof-of-Concepts
- Rapidly prototype end-to-end Computer Vision applications to validate product ideas before committing to a full build
- Translate prototype learnings into production architecture decisions
Infrastructure and Quality
- Ship production systems with containerization (Docker, Kubernetes), CI/CD (GitHub Actions), and monitoring (Prometheus, Grafana)
- Establish engineering quality through testing (pytest), observability, and clean architecture
What You Bring
Must Have
- Minimum 4 years building Computer Vision systems, with a mix of end-to-end ownership: you have taken CV models from raw data through training and evaluation into a production deployment that real users or systems depend on, rather than prototypes that stopped at a benchmark
- Hands-on model work: you have trained, fine-tuned, and evaluated your own CV models (classification, detection, segmentation), not only built infrastructure around models other people trained
- Data preparation depth: labeling and annotation quality, class imbalance, augmentation strategy, and the data-quality failure modes that quietly degrade model performance
- Deep PyTorch and OpenCV experience
- Strong Python with production-grade API development (FastAPI)
- Model packaging and serving in production (ONNX Runtime, TorchServe, or similar)
- Deployment and operations: Docker, CI/CD (GitHub Actions), and monitoring of live models
- Solid system design and architecture skills
Nice to Have
- GPU infrastructure: CUDA and cuDNN management, multi-GPU training on Ubuntu servers
- Inference optimization beyond ONNX: TensorRT, OpenVINO, Triton, or edge runtimes
- Kubernetes at production scale
- Experiment tracking and automated model evaluation (MLflow or Weights & Biases)
- Classical CV: geometry and calibration, tracking
What We Offer
- Hybrid work setup: about 2 to 3 days in office per week
- Startup feel with enterprise resources, in an international team with backgrounds from Lazada, Gojek, and IBM
- Low-bureaucracy, high-impact environment where your models and code go into real deployments
- Direct collaboration with AI researchers and computer vision scientists
- Culture of experimentation, self-development, and knowledge sharing
