Artificial Intelligence Researcher

The NLP Core team at Viettel AI researches and develops large-scale foundation models, with a long-term mission to build sovereign AI capabilities for Vietnam.Our team works across the full foundation-model stack, from large-scale data and model training to post-training, reinforcement learning, reasoning, long-context modeling, tool use, evaluation, and deployment.Following our recent work on large-scale LLM post-training, we are expanding our research toward the next generation of foundation models, with a particular focus on:Advanced Reasoning · Multimodal Intelligence · Agentic LLMsWe are looking for talented Senior and Principal AI Researchers who are excited about solving fundamental research problems and turning those ideas into large-scale, real-world AI systems.What You Will Work OnYou will research and develop core technologies for future generations of Viettel AI foundation models. Depending on your research background, you may focus deeply on one or more areas:Advanced Reasoning & Reinforcement Learning. Develop methods that improve mathematical, scientific, coding, planning, and long-horizon reasoning. Explore reinforcement learning for LLMs, verifiable rewards, reward design, policy optimization, online learning, distillation, inference-time reasoning, self-improvement, and scalable generation of reasoning data.Multimodal Foundation Models. Extend foundation models beyond text toward vision, documents, diagrams, screenshots, audio, and other modalities. Research multimodal pre-training and post-training, cross-modal reasoning, multimodal reinforcement learning, and models capable of reasoning and acting across different modalities.Agentic LLMs. Develop models that can interact effectively with tools and environments: planning, search, retrieval, function calling, memory, task decomposition, context management, error recovery, and long-horizon task execution. A major research question is how to move from models that primarily generate responses toward models that can reason, act, observe, and adapt.Long-Context & Memory. Research efficient long-context training and inference, context compression and management, retrieval-augmented reasoning, persistent memory, and reasoning over large document collections or extended interaction trajectories.Model Architecture & Efficiency. Explore Mixture-of-Experts architectures, efficient attention, speculative decoding, multi-token prediction, inference-time scaling, model compression, and other techniques that improve the capability, training efficiency, and serving efficiency of large foundation models.Data & Evaluation. Develop scalable pipelines for high-quality training data, synthetic reasoning data, agent trajectories, multimodal data, verifiers and reward signals. Design rigorous evaluations that measure not only final-answer accuracy but also reasoning, tool use, planning, robustness, and long-horizon task completion.What You Will DoThis is a research-oriented role, where you will own challenging problems from research idea to large-scale validation.You will formulate research hypotheses, design algorithms, implement prototypes, conduct controlled experiments and ablation studies, train and evaluate large foundation models, and analyze both successes and failure modes.You will work with research engineers and infrastructure teams to scale promising ideas from small experiments to large-model training and deployment.You will also contribute to the team's longer-term research agenda by studying frontier research, identifying important open problems, proposing new approaches, and communicating results through technical reports, patents, open-source contributions, and publications at leading AI conferences.At the Principal level, you will additionally help define research directions, lead major research projects, mentor researchers, and contribute to the technical roadmap for future generations of Viettel AI foundation models.What We Are Looking ForRequired qualificationsMaster's or PhD in Computer Science, Artificial Intelligence, Machine Learning, NLP, Computer Vision, Mathematics, or a related field; or equivalent research experience.Strong understanding of modern Transformers, LLMs, and foundation models.Strong research ability: identifying meaningful problems, developing hypotheses, designing rigorous experiments, and analyzing results.Strong programming skills, particularly Python and PyTorch.Hands-on experience training, fine-tuning, or conducting research on large neural models.Ability to understand, reproduce, critique, and extend recent research papers.Deep expertise in at least one of the following:LLM reasoning and reinforcement learningMultimodal foundation models / VLMsAgentic LLMs and tool useLLM post-training and alignmentModel architecture and training efficiencyLarge-scale distributed trainingWe do not expect candidates to be experts in all of these areas. We value researchers with deep expertise in one area and the ability and curiosity to collaborate across the broader foundation-model stack.Strongly PreferredExperience with one or more of the following is a strong advantage: large-scale SFT or RL training; RLVR, GRPO or related policy-optimization algorithms; reasoning models; multimodal LLM/VLM training; agent and tool-use environments; synthetic-data generation; reward models and verifiers; Mixture-of-Experts; long-context modeling; or distributed training frameworks such as Megatron-LM, NeMo, DeepSpeed, FSDP, or similar systems.Publications at NeurIPS, ICML, ICLR, ACL, EMNLP, NAACL, CVPR, ICCV, ECCV, AAAI, or comparable venues are a plus, but demonstrated research ability and technical impact matter more than publication count alone.Why Join UsThis is not primarily an API integration, prompt-engineering, or application-development role.You will work on the core intelligence of foundation models—training models, developing new algorithms, constructing learning environments, designing data and rewards, running large-scale experiments, and understanding why models succeed or fail.You will have the opportunity to work on research problems at the intersection of:Reasoning × Reinforcement Learning × Multimodality × Agents × Large-Scale Trainingand see successful research move from an idea or experiment into models and systems used in real-world applications.Our Research MissionOur long-term goal is to build world-class sovereign foundation-model technology for Vietnam: models with deep Vietnamese language and cultural understanding while advancing general reasoning, coding, scientific intelligence, multimodal understanding, and agentic capabilities.We believe this requires developing expertise across the entire model lifecycle:Data → Pre-training → Post-training → Reinforcement Learning → Reasoning → Multimodality → Agents → Inference → Evaluation → DeploymentWe are looking for researchers who want to help build that capability.