NLP/LLM Engineer

About Position:
We are looking for a highly skilled NLP/LLM Engineer to join our team and work on cutting-edge large language model (LLM) applications. The ideal candidate will have hands-on experience in prompt engineering, LLM training and fine-tuning, and model/chatbot evaluation.
You should be able to leverage state-of-the-art knowledge and recent research in NLP/LLMs to build and optimise intelligent conversational agents, while also developing scalable pipelines for model evaluation.


Responsibilities:

- Prompt Engineering

• Design, optimise, and evaluate prompts for various NLP/LLM use cases.
• Apply advanced prompting techniques such as self-consistency, chain-of-thought (CoT), few-shot, zero-shot, retrieval-augmented prompting, etc.
• Develop systematic prompt evaluation frameworks and automate prompt optimisation using both open-source and API-based models.
• Experiment with guardrails, role-based prompting, and output formatting to ensure reliability and safety in chatbot responses.

- Model Training & Fine-Tuning

• Develop and train LLMs using PyTorch, Hugging Face, Unsloth, and vLLM.
• Implement multi-GPU/distributed training with DeepSpeed, FSDP, or Accelerate.
• Fine-tune models with Supervised Fine-Tuning (SFT) and Reinforcement Learning ( PPO, DPO, GRPO, etc.).
• Apply parameter-efficient fine-tuning (PEFT) methods (LoRA/QLoRA, IA3, etc.) for efficiency.
• Familiarity with mixed precision, FlashAttention, quantisation, gradient checkpointing, optimiser sharding (ZeRO), and other advanced training techniques.
• Build and maintain data preprocessing pipelines for large-scale NLP training datasets.
• Track and document model training experiments for reproducibility (e.g., using Weights & Biases or MLflow).

- Model & Chatbot Evaluation

• Develop evaluation pipelines for both general LLMs and RAG-based chatbots.
• Apply automatic evaluation metrics such as BLEU, ROUGE, METEOR, BERTScore, perplexity, etc.
• Use LLM-as-a-judge methods (both open-source and closed-source via APIs) for qualitative evaluation.
• Design user simulation frameworks for stress-testing chatbots in real-world scenarios.
• Build modular and scalable evaluation pipelines to benchmark model accuracy, robustness, safety, and latency.

- Software Engineering Practices

• Use Git for version control, branching workflows, and collaborative development.
• (Optional but preferred) Experience with modern software engineering tools and practices, including Docker, CI/CD pipelines, and Kubernetes.

Requirements:

• Strong background in Natural Language Processing (NLP) and Large Language Models (LLMs).
• Proven experience in prompt engineering with real-world applications.
• Hands-on experience in training and fine-tuning large models using Hugging Face, PyTorch, and distributed training setups.
• Practical knowledge of applying PEFT methods, SFT, and RLHF/RLVR on LLMs.
• Experience in designing evaluation frameworks for LLMs and chatbots.
• Proficiency with Git for collaborative development.
• Strong problem-solving skills, ability to work reliably, and encouragement to collaborate in a team environment.
• (Preferred) Familiarity with Docker, CI/CD pipelines, and Kubernetes.

What We Offer
A dynamic, fast-paced environment where your ideas directly impact real-world products.
A collaborative team culture that values innovation, technical excellence, and continuous learning.
The chance to help shape the future at MCINEXT.