Senior LLM / Multimodal AI Engineer

About the role

NCS, an AI tech services company and subsidiary of the Singtel Group, is hiring a Senior LLM / Multimodal AI Engineer in Singapore. The engineer will build and deploy large-scale Vision-Language Models (VLMs) and Large Language Models (LLMs) that process billions of user-generated content and advertising videos, turning research into industrial-scale recommendation and advertising systems.

Key responsibilities

• Data curation and multimodal pre-training: design automated pipelines integrating video frames, images, OCR, ASR and text, and drive LLM-assisted annotation and synthetic data generation
• Model training and fine-tuning: reproduce and optimise foundation models such as LLaVA and Qwen-VL, run supervised fine-tuning and post-training for domains like gaming, movies, automotive and advertising, and use PEFT, LoRA and quantization-aware training
• Multimodal RAG and knowledge bases: build hierarchical RAG pipelines, vector search indexes (e.g. FAISS) and hybrid search, re-ranking and query expansion strategies
• Inference optimisation and deployment: serve VLMs and LLMs with vLLM, TensorRT-LLM, Triton and ONNX Runtime, and tune latency, cost and quality
• Evaluation and agent workflows: set up online/offline benchmarks, contribute to tag platform agents and run LLM-as-a-judge evaluations

Required profile

• 4+ years of hands-on experience in machine learning system development, focused on NLP, computer vision, LLMs or multimodal AI
• Proven track record of driving end-to-end ML initiatives from research to high-throughput production deployment
• Preferred: experience in recommender systems, advertising technology or content moderation, RLHF or preference optimisation (DPO, GRPO), published research, and cloud ML platforms (AWS SageMaker, Azure ML, Google Vertex AI)

Required skills

• PyTorch, Hugging Face ecosystem, DeepSpeed, PEFT, LlamaFactory
• Python, SQL, Spark, Linux and Git
• FAISS and multimodal RAG, and serving toolkits such as vLLM, TensorRT-LLM and Triton