Multimodal AI Engineer (Robotics & Foundation Models)

Asiacruit BPO, Inc.

15h•Hybrid

About The Position

Asiacruit is seeking a Multimodal AI Engineer to develop and adapt Vision-Language-Action (VLA) foundation models for autonomous robotic systems. The role works across multimodal perception, foundation models, and robotic learning, enabling AI systems to interpret visual inputs, plan complex tasks, and translate instructions into physical actions. The work includes fine-tuning multimodal models, developing multi-camera perception pipelines, implementing action-generation architectures, and optimizing models for deployment on edge AI platforms. The engineer will collaborate with AI researchers and robotics engineers to turn research concepts into practical, high-performance AI capabilities.

Requirements

Master's degree or Ph.D. in Computer Science, Artificial Intelligence, Robotics, Machine Learning, Computer Vision, or a related field.
Approximately 3–5 years of relevant experience in multimodal AI, deep learning, computer vision, foundation models, or robotic learning.
Strong programming skills in Python and hands-on experience with PyTorch.
Experience developing, training, or fine-tuning Transformer-based architectures, Vision Transformers, or multimodal deep learning models.
Understanding of vision-language models, multimodal representation learning, and the integration of visual information into AI decision-making systems.
Practical experience with parameter-efficient fine-tuning methods, such as LoRA or other PEFT techniques.
Familiarity with robotic learning and action-generation approaches, including ACT, Diffusion Policy, imitation learning, or reinforcement learning.
Experience with multi-camera perception, visual feature fusion, or related computer vision pipelines.
Understanding of model profiling, pruning, quantization, or other model optimization techniques.
Strong analytical and problem-solving skills, with the ability to conduct experiments and translate research findings into practical engineering solutions.
Willingness to work in a hybrid arrangement and be available for on-site work at a designated research location, if required for future projects.

Nice To Haves

Familiarity with NVIDIA DeepStream, TensorRT, or GPU-accelerated inference.
Knowledge of embeddings, vector search, and multimodal data processing.

Responsibilities

Develop, fine-tune, and adapt Vision Transformers (ViTs) and multimodal foundation models for robotic perception and action generation.
Implement action-chunking architectures, including Action Chunking with Transformers (ACT) and Diffusion Policy, to support long-horizon robotic task planning.
Build multi-camera fusion pipelines that combine visual inputs into coherent representations for VLA perception and decision-making.
Develop and optimize multimodal feature extraction pipelines that integrate visual perception with downstream robotic planning and action-generation systems.
Integrate computer vision capabilities, including face recognition and perimeter threat assessment, into multimodal AI workflows.
Apply parameter-efficient fine-tuning (PEFT) techniques, including LoRA, to adapt foundation models for specific robotic applications.
Work with Transformer architectures, visual embeddings, and vector search techniques to improve multimodal representation and information retrieval.
Profile and optimize multimodal models through pruning and other model compression techniques to improve inference efficiency.
Prepare and optimize model weights for downstream deployment using NVIDIA TensorRT and related inference optimization tools.
Evaluate model performance through systematic experiments, benchmarking, and iterative improvements.
Collaborate with researchers and engineering teams across VLA, robotics controls, and edge AI to support end-to-end system development.