Job Type: Full-Time

ABOUT THE ROLE
Fine-tune large language models for low-resource languages, with a focus on Nepali, and help build foundational language AI for underserved communities.

KEY RESPONSIBILITIES

• Fine-tune open-source LLMs (LLaMA, Mistral, Qwen, etc.) using LoRA/QLoRA, SFT, and instruction tuning.
• Curate, clean, and augment training data for low-resource languages.
• Build and adapt tokenizers for Devanagari and other target scripts.
• Design evaluation benchmarks and run rigorous experiments.
• Optimize and deploy models for production inference.

REQUIRED SKILLS & EXPERIENCE

• 2–4 years of NLP / deep learning experience.
• Hands-on LLM fine-tuning experience (must discuss specific projects).
• Strong Python and PyTorch + Hugging Face stack (Transformers, PEFT, TRL).
• Solid grasp of transformer architectures and modern training techniques.
• Experience working with Devanagari or other non-Latin scripts.

NICE TO HAVE

• Open-source contributions or research in low-resource NLP.
• Experience with distributed training (DeepSpeed, FSDP) and inference frameworks (vLLM, TGI).
• Familiarity with RLHF/DPO and RAG architectures.

WHAT WE OFFER

• Collaborative and learning-driven work culture
• Career growth and professional development
• Competitive salary and benefits