LLM Research Engineer

Novo AI; Hong Kong; On-site; HKD 420,000 - 700,000; Full time

Job summary

Novo AI in Hong Kong is seeking researchers to advance applied ML for real-world claims processing. You will read patient records, code with official rules, and build explainable systems that justify every decision.

Collaborate on retrieval over rulebooks, design evidence-linked outputs, run benchmarks with clinicians, and document methods so others can reproduce results.

Qualifications

• Bachelor's degree in a science, technology, engineering or mathematics discipline.
• Solid programming ability in Python or TypeScript.
• Working knowledge of machine learning or natural language processing fundamentals.
• Comfort with experiments and measuring results honestly.
• Patience with detail and reading structured rules.
• Legally permitted to work in Hong Kong.

Responsibilities

• Research how large language models can read patient records and assign standard medical codes using official rules.
• Build retrieval over versioned rulebooks so grounding changes answers.
• Design outputs where each code shows source sentence and justification with a deterministic checker.
• Build benchmarks, locked holdout sets, and measure against published ML results and human baselines.
• Design escalation logic to flag unclear evidence or rules rather than guessing.
• Run error analysis with clinical reviewers and document methods for reproducibility.

Skills

Python

TypeScript

ML/NLP fundamentals

Education

Bachelor's degree in STEM

Tools

Git

VS Code

Job description

• Applied research on a genuinely unsolved problem, measured against published baselines rather than demos.
• Build systems that explain themselves. Every decision has to carry its evidence and the rule behind it.
• Learn from an ex-Google applied machine learning engineer and from practising clinicians.
• Your results change how insurers handle real claims.

What you'll work on

• Research how large language models can read a patient record and assign standard medical classification codes by applying the official coding rules, rather than pattern-matching from training data.
• Build retrieval over versioned rulebooks so the system reasons from current guideline text, and measure whether that grounding changes the answer.
• Design evidence-linked outputs, where every code carries the source sentence and the rule that justifies it, and a deterministic checker rejects anything unsupported.
• Build benchmarks and locked holdout sets, and measure accuracy against published machine learning results and human specialist baselines.
• Design escalation logic that flags cases where the evidence or the rule is unclear, instead of guessing.
• Run error analysis with clinical reviewers, and write up methods and results so another engineer can reproduce them.

Requirements

• Bachelor's degree in a science, technology, engineering or mathematics discipline.
• Solid programming ability in Python or TypeScript.
• Working knowledge of machine learning or natural language processing fundamentals, from coursework, research or personal projects.
• Comfort with experiments. You can form a hypothesis, design a measurement, and read the result honestly.
• Patience with detail. Much of this work is careful reading of structured rules and taxonomies.
• Legally permitted to work in Hong Kong.

Nice to have

• Projects or coursework in natural language processing, information retrieval or retrieval-augmented generation.
• Experience evaluating LLM systems, including building test sets and scoring rubrics.