Lead Data Scientist

About Us We are building transformer models focused on multilingual translation and custom transformer architectures.
Our team works on large-scale NLP and transformer-based models with a strong focus on research, experimentation, model optimization, and production-grade AI systems.
We are looking for a highly technical Data Science Lead who can drive end-to-end model development, mentor engineers, and lead the evolution of our AI stack.
This is a core engineering and research role not a prompt engineering or API integration role.
Role Overview As a Data Science Lead, you will lead the design, training, fine-tuning, evaluation, and optimization of transformer-based NLP systems and LLMs.
You will work closely with engineering and research teams to build scalable AI models from the ground up and establish technical direction for the organization.
You should be comfortable working deeply with model internals, training pipelines, datasets, tokenization strategies, and distributed GPU-based training environments.
Key Responsibilities Lead the architecture and development of transformer-based AI systems Drive technical direction for NLP, LLM, and multilingual AI initiatives Mentor and guide junior ML engineers and researchers Train and fine-tune transformer models on large-scale custom datasets Work on Seq2 Seq / encoder-decoder architectures for translation and text generation Optimize model performance using Lo RA, QLo RA, PEFT, quantization, and distributed training techniques Design tokenization pipelines using BPE / Sentence Piece Evaluate models using BLEU, Chrf, accuracy, and custom benchmarks Collaborate with platform teams for GPU infrastructure and scalable deployments Required Qualifications 6 years of experience in Data Science / Machine Learning / NLP Strong hands-on expertise in Transformers, LLMs, Seq2 Seq Architectures, and Attention Mechanisms Proven experience in model training and fine-tuning using custom datasets Hands-on experience with Hugging Face, Py Torch, or Tensor Flow Vector Databases: Pinecone, Milvus – for embeddings and semantic search in translation or LLM applications.