$19–23/hr · Mercor · Part time, 7 hours a week
A doctoral scientist who writes Indonesian scientific questions and evaluates how AI models handle chemistry and biology topics safely, without requiring machine learning background.
What you would do
- Compose detailed scientific scenarios and inquiries in Indonesian covering areas such as synthesis methods, laboratory safety, and biological techniques
- Rate model-generated responses against scientific standards, evaluating factual correctness and methodological soundness
- Annotate outputs to flag sensitive information handling, potential dual-use applications, or inappropriate guidance on hazardous procedures
- Apply structured classification systems to document model behavior across hundreds of chemistry and biology test cases
Who they want
- Completed or ongoing doctoral degree in chemistry, biology, molecular sciences, or closely related field
- Native or near-native command of Indonesian alongside professional-level written and spoken English
- Hands-on experience with laboratory methods, computational tools, or both in your specific research domain
- Mature judgment regarding scientific ethics and the implications of information disclosure in sensitive technical fields
Main skills
What the interview asks about
1.Indonesian technical terminology precision
Model evaluation depends on you formulating questions that test precise algorithmic understanding; weak Indonesian scientific terminology will either mislead the model or fail to probe its capabilities authentically.
For example: “Your team is evaluating chemistry models on synthetic route design. Compose a prompt in Indonesian asking about reagent selection for a multi-step organic synthesis, ensuring the language captures the precise reactions you want the model to consider.”
2.Distinguishing scientific accuracy from information hazard
A response can be scientifically correct yet represent an unacceptable information disclosure; your judgment determines whether an output passes or fails safety review.
For example: “A model describes precursor chemicals and suppliers for a synthesis. It's published-accurate but potentially accelerates dangerous work. How would you categorize this for engineers?”
3.Deep domain-specific knowledge assessment
You must recognize not just right answers but appropriate context, current best practices, and limitations that only a trained researcher would know; shallow familiarity with scientific topics will miss significant flaws.
For example: “A model addresses microbiology safety protocols in Indonesian but omits containment procedures standard in your field. Is this a model gap or acceptable brevity?”
4.Consistency in applying safety guidelines
Annotation standards must hold across hundreds of reviews; inconsistent application corrupts the training signal and risks either over-filtering benign content or under-catching actual hazards.
For example: “You've reviewed 150 chemistry responses using the safety rubric. You notice your threshold for flagging synthetic pathway information has drifted over time. Describe how you would recalibrate and what documentation you would keep to maintain consistency.”
5.Cross-lingual concept bridging
The engineering team operates primarily in English; your ability to explain Indonesian technical nuances and why a model's response misses cultural or linguistic context determines whether they can fix it.
For example: “A model's Indonesian response reflects outdated safety standards from another region instead of current Indonesian practice. How would you explain this gap to engineers?”
A task you may get
Review ten Indonesian chemistry prompts with model responses. Rate each on accuracy, safety, and completeness, then summarize your highest-risk and top-performing cases.
How to prepare
- Review recent literature in your research domain to identify how scientific consensus has shifted in the last 2-3 years, particularly around safety and responsible disclosure
- Compile a personal reference of technical terminology you use regularly in Indonesian and English to ensure precision during real-time evaluation
- Study examples of dual-use research concerns in chemistry and biology to develop pattern recognition for risk flagging
The facts
- Pay
- $19–23/hr
- Hours
- Part time, 7 hours a week
- Where
- Remote · Remote — Southeast Asia preferred
- Field
- Life, Physical, and Social Science
- Posted
- 9/4/2026
- Places left
- 1
We wrote this page from the public Mercor listing. It may be out of date, so read the full posting before you apply.