Work / 05
Domain expertise inside AI systems
Evaluating AI-generated environmental-management content against real-world practice
- Organisation
- Labelbox / Alignerr
- Role
- Environmental Management Scientist (AI training)
- Years
- 2026–present
- Location
- Remote (Portugal)
39.820° N · 7.490° W - Scale
- global
Challenge
Large language models write fluent environmental-management advice. Fluent is not the same as correct: mitigation hierarchies, EIA logic, land-use constraints and regulatory frameworks have structure that a model can imitate without respecting.
Data
- AI-generated content on land-use planning, impact mitigation and environmental decision-making
- Environmental frameworks and practice as the reference standard
Intelligence
- Structured evaluation of outputs against domain frameworks
- Corrections and rankings that feed back into model training
Technology
- AI model evaluation
- Deterministic LLM prompting
- Environmental frameworks
Outcome
- Model outputs that align with how environmental decisions are actually made
- A working understanding of where AI helps environmental analysis — and where it needs a specialist in the loop
Why this belongs next to the modelling work
The same skill that makes a niche model trustworthy — knowing what the data can and cannot support — is what makes an AI evaluation useful. The work is less about labelling and more about encoding judgement.
TODO — Nuno: if you can share (non-confidential) examples of the kinds of tasks reviewed, one paragraph here would make this concrete.