AI Research Engineer (Post-Training)

H2 Games; Singapore; On-site; SGD 120,000 - 180,000; Full time

Job summary

H2 Games is seeking a Machine Learning/AI Research Engineer to adapt, evaluate, and productionize multimodal foundation models for image assets and 2D animation. You will improve image generation/editing, visual understanding, style consistency, and asset-quality assessment while enabling animation support and content-quality evaluation.

The role requires turning research prototypes into reproducible production systems, collaborating with product, design, and engineering teams, and building

Qualifications

• More than 1 year of experience in machine learning, computer vision, multimodal AI, generative AI, or applied AI research.
• Strong PyTorch skills and experience with model training, fine-tuning, evaluation, deployment, or inference optimization.
• Hands-on experience with SFT, LoRA/PEFT, DPO/RLHF/RLAIF, reward modeling, model evaluation, or continual learning.

Responsibilities

• Evaluate and select visual, multimodal, and image-generation foundation models; establish measurable baselines and iteration roadmaps.
• Run supervised fine-tuning, LoRA/PEFT, preference optimization, continual adaptation, distillation, and other post-training methods using proprietary data.
• Develop capabilities for image generation and editing, visual understanding, automatic tagging, style recognition, character consistency, asset-quality assessment, and near-duplicate detection.
• Develop 2D-animation capabilities including part understanding, layer-extraction assistance, pose and skeleton inference, rigging assistance, motion transfer, animation generation, and quality assessment.
• Combine model outputs with structured constraints, deterministic validation, safety policies, and human-review workflows to improve reliability, editability, and publishability.
• Build offline and human evaluation systems to measure generation quality, edit success rate, style consistency, first-pass acceptance, user satisfaction, latency, and inference cost.
• Partner with the AI Data Engineer to turn model failures, artist corrections, user feedback, and moderation results into reusable training and evaluation data.
• Work with product, design, and engineer teams to deliver production services, batch workflows, and creator tools.
• Maintain reproducible training, evaluation, inference, monitoring, and model-versioning workflows.

Skills

PyTorch

Model training

Fine-tuning

Multimodal AI

Evaluation

Production systems

Cross-functional collaboration

Education

Master's degree in ML or related field

Tools

LoRA/PEFT

SFT

DPO/RLHF/RLAIF

Diffusion models

Job description

We are building multimodal AI capabilities for creative-content production across image assets and 2D animation. You will adapt, evaluate, and productionize visual, multimodal, and generative foundation models to improve image generation and editing, visual understanding, style and character consistency, animation assistance, and content-quality evaluation.

Responsibilities

• Evaluate and select visual, multimodal, and image-generation foundation models; establish measurable baselines and iteration roadmaps.
• Run supervised fine-tuning, LoRA/PEFT, preference optimization, continual adaptation, distillation, and other post-training methods using proprietary data.
• Develop capabilities for image generation and editing, visual understanding, automatic tagging, style recognition, character consistency, asset-quality assessment, and near-duplicate detection.
• Develop 2D-animation capabilities including part understanding, layer-extraction assistance, pose and skeleton inference, rigging assistance, motion transfer, animation generation, and quality assessment.
• Combine model outputs with structured constraints, deterministic validation, safety policies, and human-review workflows to improve reliability, editability, and publishability.
• Build offline and human evaluation systems to measure generation quality, edit success rate, style consistency, first-pass acceptance, user satisfaction, latency, and inference cost.
• Partner with the AI Data Engineer to turn model failures, artist corrections, user feedback, and moderation results into reusable training and evaluation data.
• Work with product, design, and engineer teams to deliver production services, batch workflows, and creator tools.
• Maintain reproducible training, evaluation, inference, monitoring, and model-versioning workflows.

Qualifications

• More than 1 year of experience in machine learning, computer vision, multimodal AI, generative AI, or applied AI research.
• Strong PyTorch skills and experience with model training, fine-tuning, evaluation, deployment, or inference optimization.
• Hands-on experience with one or more of SFT, LoRA/PEFT, DPO/RLHF/RLAIF, reward modeling, model evaluation, or continual learning.
• Expertise in at least one relevant area: image understanding, image generation/editing, segmentation, pose estimation, diffusion models, vision-language models, or video/animation generation.
• Ability to turn research prototypes into reproducible, observable, and maintainable production systems.
• Strong experimental-design, data-analysis, and cross-functional collaboration skills.

Preferred Qualifications

• Experience shipping AI for image-asset platforms, design tools, advertising creative, game art, or creator products.
• Experience with 2D character animation, Spine/Live2D, game assets, computer graphics, rigging, or kinematics.
• Familiarity with structured generation, JSON/schema constraints, geometry constraints, content safety, or graphics-asset validation.
• Experience with distributed training, GPU optimization, quantization, inference acceleration, or cost optimization.
• Experience shipping generative-AI products, publishing research, or contributing to open-source projects.