AI Engineer – Computer Vision & Vision-Language Models

SECOND TALENT SG PTE. LTD.; Singapore; On-site; SGD 90,000 - 150,000; Full time

Job summary

SECOND TALENT SG PTE. LTD. is seeking a machine learning engineer to own and modernise a production defect-detection and classification model stack, focusing on building reliable GPU-accelerated pipelines on AWS.

You will migrate legacy computer-vision workloads, design evaluations for vision-language models, and contribute to photogrammetry and 3D reconstruction workflows using drone imagery. The role requires hands-on PyTorch/CUDA, Python, and Docker, with experience in S3/SageMaker and

Qualifications

• 2+ years of ML/computer-vision engineering with models in production.
• 2+ years running training/inference on AWS (EC2 GPU and/or SageMaker).
• Strong PyTorch and CUDA/GPU engineering background.
• Recent hands-on experience with vision-language or multimodal models.
• Python and boto3 automation (S3, EC2, SSM, Lambda).
• Model serving experience (TorchServe, Docker on EC2).
• Comfortable administering both Linux and Windows GPU hosts.
• Experience with photogrammetry, 3D reconstruction (e.g.NeRF/Gaussian Splatting) or drone/geospatial imagery is a strong plus

Responsibilities

• Own and modernise the production defect-detection andclassification model stack.
• Migrate legacy computer-vision workloads to supportedruntimes and modern serving infrastructure.
• Design and run a structured evaluation of managed vs.self-hosted vision-language models (VLMs).
• Operate distributed inference and GPU batch processingpipelines on AWS.
• Contribute computer-vision stages to photogrammetry and3D reconstruction pipelines.
• Build dataset curation, annotation and evaluationworkflows over large-scale imagery data.
• Establish model evaluation benchmarks and versionedground-truth datasets.

Skills

ML in production

AWS GPU

PyTorch

Vision-language

Python automation

TorchServe/Docker

Linux/Windows GPU

Job description

Our client is a Singapore-basedAI company that uses computer vision to analyse drone and telephoto imagery of building exteriors, detecting defects such as cracks, spalling and corrosion for building-compliance inspections. Their production environment runs on AWSwith a GPU compute fleet supporting photogrammetry, 3D reconstruction and AImodel training/serving.

What You’ll Do

• Own and modernise the production defect-detection andclassification model stack
• Migrate legacy computer-vision workloads to supportedruntimes and modern serving infrastructure
• Design and run a structured evaluation of managed vs.self-hosted vision-language models (VLMs)
• Operate distributed inference and GPU batch processingpipelines on AWS
• Contribute computer-vision stages to photogrammetry and3D reconstruction pipelines
• Build dataset curation, annotation and evaluationworkflows over large-scale imagery data
• Establish model evaluation benchmarks and versionedground-truth datasets

Core Key Skills

• 2+ years of ML/computer-vision engineering with modelsin production
• 2+ years running training/inference on AWS (EC2 GPUand/or SageMaker)
• Strong PyTorch and CUDA/GPU engineering background
• Recent hands-on experience with vision-language ormultimodal models
• Python and boto3 automation (S3, EC2, SSM, Lambda)
• Model serving experience (e.g. TorchServe, Docker onEC2)
• Comfortable administering both Linux and Windows GPUhosts
• Experience with photogrammetry, 3D reconstruction (e.g.NeRF/Gaussian Splatting) or drone/geospatial imagery is a strong plus

Nice to Have

• GIS/geospatial tooling familiarity
• Experience in building-inspection or similar regulateddomains
• Image restoration/enhancement or thermal imageryanalysis experience