ABOUT THE ROLE
We are looking for a specialist with solid foundations in natural language or speech processing, combined with hands-on experience training models and putting them to work in real products. You will own AI components of our products end to end - from trying out approaches, through training and evaluation, to running them in production - with the rest of the team alongside you on the harder infrastructure questions.
The role suits someone who enjoys deep technical work, makes decisions grounded in measurement, and wants to see their research turned into products that people actually use.
KEY RESPONSIBILITIES
• Research and evaluate candidate approaches for the NLP, LLM and Speech problems our products address, and make the case for the one worth building on.
• Build, train and fine-tune models on real-world data, and take ownership of output quality against clearly defined metrics.
• Build model-driven systems such as RAG and AI agents, together with the evaluation tooling that keeps their quality measurable.
• Help optimize latency, throughput and operating cost so that systems stay stable as traffic grows.
• Work alongside data, engineering and product teams to move models from experiment into production.
• Stay current with new research and tooling, and share what is worth adopting with the rest of the team.

• At least 2 years of experience as an AI/ML Engineer, Research Engineer or in an equivalent role, with hands-on depth in at least one of NLP, LLM or Speech (ASR/TTS).
• Solid foundations in machine learning and deep learning, a working understanding of the Transformer architecture, and enough comfort with papers to take an idea from one and put it to work in code.
• Hands-on experience training or fine-tuning models end to end: preparing data, running experiments, choosing how quality is measured, and improving results on the basis of what you measured.
• Experience building at least one LLM-based system in practice - RAG, an agent, or something comparable - including how you went about evaluating whether it worked.
• An understanding that a model has to run in production: that latency, throughput and cost are part of the problem rather than someone else's afterthought.
• Proficiency in Python and hands-on experience with PyTorch or TensorFlow.
• Sound problem-solving, a collaborative way of working, and enough English to be comfortable with technical material.
PREFERRED QUALIFICATIONS
• Experience fine-tuning large language models (SFT, LoRA/QLoRA; RLHF/DPO a plus), including on multi-GPU setups.
• Experience with agent systems - tool calling, multi-step workflows, and the orchestration harness around them.
• Experience serving models under real traffic: high-performance stacks (vLLM, SGLang, Triton, RayServe or equivalent), batching and caching, or control of latency and infrastructure cost.
• Experience operating models in production under an MLOps discipline: monitoring, catching quality regressions, and shipping new versions.
• Publications, open-source contributions, or strong results in recognized competitions.
WHAT WE OFFER
• Compensation of up to VND 40,000,000 per month, negotiable according to capability and reviewed periodically against results.
• Bonus scheme, insurance and leave entitlement
• Compute and data resources dedicated to research and development work.
• An open working environment with genuine technical autonomy.