LLM Direction Algorithm Engineer

X STAR TECHNOLOGY PTE. LTD.; Singapore; On-site; SGD 70,000 - 100,000; Full time

Job summary

A leading technology firm in Singapore is searching for an LLM Engineer to design and implement algorithms that steer LLM output. Responsibilities include optimizing inference architecture, collaborating with interdisciplinary teams, and investigating advanced techniques in LLM alignment. The ideal candidate holds an MS or PhD in relevant fields and possesses strong Python programming skills and experience with ML frameworks. Strong analytical and software engineering skills are essential for success in this role.

Qualifications

• Strong programming skills in Python and experience with ML frameworks.
• Solid understanding of NLP fundamentals and experience in various NLP tasks.
• Experience with fine-tuning LLMs or related alignment work.

Responsibilities

• Design and implement algorithms for controlling LLM output.
• Optimize inference architecture and serve models at scale.
• Collaborate with various teams to integrate LLM capabilities.

Skills

Strong programming skills (Python)

Experience with ML frameworks (PyTorch, TensorFlow)

Understanding of NLP fundamentals

Analytical/problem-solving skills

End-to-end system building

Education

MS or PhD in Computer Science, Machine Learning, NLP, AI

Tools

Transformer-based LLMs (LLaMA, GPT-4, Falcon, etc.)

Job description

Design and implement algorithms that steer or direct LLM output such as: controlling tone/voice/persona, enforcing constraints (e.g., style, accuracy, safe behavior), prompting/control flow, tool‑use by the model, and multi‑step reasoning workflows.

Work on post‑training processes such as supervised fine‑tuning (SFT), reinforcement learning from human feedback (RLHF), reward modelling and alignment techniques to shape model behavior. For example, one job listing describes “advanced post‑training of large language models (e.g. SFT, RLHF/RLAIF…)”.

Build and maintain evaluation & benchmark pipelines for behavior: e.g., factuality, bias/ethics, latency, throughput, tool‑calls, JSON‑schema compliance, multi‑round interaction, hallucination detection. For instance: “Build offline and live eval pipelines for alignment, factuality, grounding, and hallucinations.”

Optimize inference architecture and serve models at scale: e.g., context window management, checkpoint routing, efficient serving, adapter/LoRA training, distributed training/inference frameworks. As one role states: “Design, deploy, and operate Model Context Protocol (MCP) servers … manage context windows.”

Collaborate with research, product, safety, and infrastructure teams to integrate directed‑LLM capabilities into product experiences—ensuring the model’s behavior meets user intent, product goals, and safety/ethics standards.

Investigate and adopt state‑of‑the‑art techniques in LLM alignment, prompting/steering, retrieval‑augmented generation (RAG), multi‑modal integration, agentic behavior, and emergent capabilities.

What We’re Looking For / Qualifications

Required:

• MS or PhD in Computer Science, Machine Learning, NLP, AI, or equivalent experience.
• Strong programming skills (Python) and experience with ML frameworks (e.g., PyTorch, TensorFlow) as well as working knowledge of Transformer‑based LLMs (e.g., LLaMA, GPT‑4, Falcon, etc.).
• Solid understanding of NLP fundamentals (language modelling, sequence‑to‑sequence, embeddings, attention mechanisms) and experience in tasks such as question‑answering, summarisation, dialogue, retrieval.
• Experience with at least one of: fine‑tuning LLMs, RLHF, alignment work, prompt engineering, RAG setups.
• Good software engineering skills: end‑to‑end system building (data pipelines, model training, serving/inference, evaluation).
• Strong analytical/problem‑solving skills and ability to work cross‑functionally in a fast‑moving environment.

Preferred:

• Hands‑on experience with distributed training/inference frameworks: e.g., DeepSpeed, FSDP, vLLM, Triton.
• Experience with tool‑use by LLMs, function‑calling, agents, multi‑step reasoning workflows—e.g., “aligning models for reliable JSON‑schema function calls and external tool usage”.
• Familiarity with safety, bias/ethics, fairness, adversarial robustness in LLMs.
• Experience with multi‑modal (text+image+code) or agentic interfaces.
• Publications in top ML/NLP/AI conferences (ICLR, NeurIPS, ACL, EMNLP) or strong open‑source contributions.