Neural Machine Translation Consultant Full Time on the ground NEW

Sony Research India is driving cutting-edge research and development in various locations around the globe, including laboratories in Japan, the United States, Europe, and Asia. We endeavor to create new technology, products, and services while sustaining Sony Group’s diverse businesses in electronics, entertainment, and financial fields. For our research centre to blaze a trail in the latest technologies, we seek to foster the growth of a diverse pool of research and engineering talent and create a technology talent bank to drive research excellence worldwide. Sony Research India is offering outstanding career opportunities around frontline technologies such as AI and data analytics.

What we are looking for:

• We are seeking a highly motivated NMT/LLM Consultant to work on advanced LLM-based Neural Machine Translation systems, with a strong focus on model architecture design, pipelined inference/training systems, and model acceleration. The role involves close collaboration with research scientists and engineers to design scalable, efficient, and high-quality multilingual translation systems, particularly for entertainment-domain content across Indian and international languages.

Key Responsibilities:

• Research and design LLM-based NMT model architectures, including modular and extensible translation systems.
• Investigate model acceleration techniques for training and inference, including quantization, distillation, pruning, and efficient caching mechanisms.
• Design and evaluate domain-specific translation pipelines, particularly for entertainment content (e.g., subtitles, dialogues, conversational text).
• Explore low-resource and semi-supervised translation techniques.
• Contribute to evaluation methodologies, error analysis, and qualitative assessment for multilingual and domain-specific translation.
• Collaborate with research scientists and engineers to benchmark models, run large-scale experiments, and improve system throughput and latency.

Work Location:

Remote within India

Duration of the paid one year contract:

• The annual paid direct contractual tenure is extendable.
• Ideally this position will start from second week March 2026.
• The working hours are from 9:00 to 18:00 (Monday to Friday) full-time.

Essential Education:

• Ph.D. or Masters in Computer Science background or Bachelor’s with hands-on experience on Python with an applications in Language Processing.

Skills required:

Essential Skills:

• Strong background in Deep Learning and Natural Language Processing
• Familiarity with Reinforcement Learning (theoretical or applied)
• Excellent programming skills in Python, with practical experience in PyTorch
• Solid understanding of LLMs, foundation models, and modern NMT architectures
• Ability to read, implement, and extend research papers

Good to have Skills:

• Preference-based learning, reinforcement learning, or optimization for language models
• Pipelining and parallelism (data, model, tensor, and pipeline parallelism)
• LLM-based NMT improvements, implementing architectural changes, tokenizer-free models, improving and developing new finetuning approaches
• Publications or prior research exposure in top-tier conferences/journals (e.g., ICLR, ICML, NeurIPS, ACL, EMNLP, AAAI)
• Work on model architecture compression and acceleration, efficient model caching.
• Systems-oriented ML research for scalable language models

Our Values:

Dreams & Curiosity: Pioneer the future with dreams and curiosity.

Diversity: Pursue the creation of the very best by harnessing diversity and varying viewpoints.

Integrity & Sincerity: Earn the trust for Sony brand through ethical and responsible conduct.

Sustainability: Fulfil our stakeholder responsibilities through disciplined business practices.

Sony Research India is committed to equal opportunity in all its employment practices, policies and procedures and to ensuring that no worker or potential worker will receive less favourable treatment due to any characteristic protected under applicable local laws.

Source ⇲