AI Developer – LLM Model Development & Training

Location: Bengaluru

Employment type: Full-time

Joining timeline: Within 30 days

Role overview

We’re looking for a hands-on AI Developer to build and train large language models, from preparing training data to evaluating models and handing them over for deployment. You’ll develop training and fine-tuning approaches, run experiments on multi-GPU systems, and work closely with the inference team to make models accurate, efficient, and ready for production.

Key responsibilities

·      Design and run LLM pre-training, continued pre-training, and supervised fine-tuning.

·      Prepare high-quality training data through collection, cleaning, filtering, deduplication, and tokenization.

·      Run and troubleshoot distributed training across multiple GPUs; improve training throughput and reliability.

·      Evaluate models using public benchmarks, internal datasets, and regression tests.

·      Track experiments, compare results, and document model performance and training decisions.

·      Explore techniques such as parameter-efficient fine-tuning, distillation, and model compression.

·      Collaborate with the inference team to deliver deployment-ready models.

Required qualifications

·      Bachelor’s or Master’s degree in Computer Science, AI/ML, Data Science, Engineering, Mathematics, or a related field.

·      2–4 years of hands-on machine learning or deep learning experience, including at least 1 year working with transformer models or LLMs.

·      Experience training or fine-tuning an LLM on multiple GPUs.

·      Strong Python and PyTorch skills, with familiarity with Hugging Face Transformers and Datasets.

·      Understanding of transformer architecture, training workflows, optimizers, learning-rate schedules, and mixed-precision training.

·      Experience with distributed training and model evaluation.

·      Strong data processing skills and a careful, reproducible approach to experimentation.

Good to have

·      Experience with large-scale text or code data pipelines.

·      Familiarity with parameter-efficient fine-tuning, preference optimization, or synthetic data generation.

·      Experience with distributed training frameworks such as DeepSpeed, FSDP, or Megatron-LM.

·      Familiarity with evaluation tools such as lm-evaluation-harness.

·      GitHub contributions, research publications, or a portfolio of LLM projects.