NaturecodeProject
← Back to research

Efficient and sovereign computation

The infrastructure that reads the living world is itself part of the living world. Models that consume a city's worth of electricity to produce a map are not free. Data centers a continent away from the place a measurement was taken are not neutral. Our research is about taking both costs seriously — energy and sovereignty — and working on solutions that do not pretend either is solved.

What we study

  • Smaller-footprint AI. Smaller models, distilled models, sparse models, retrieval-grounded models — the work of getting useful answers without the energy budget of frontier training runs. Open methods, openly benchmarked.
  • Right-sized intelligence. Not every ecological question needs a large model. We study which problems are well served by tiny, specialized models, and which actually need scale — and we publish honest comparisons.
  • Edge-first systems. AI that runs on devices close to where the data is generated — a ranger's phone, a sensor in a river, a node in a community network. The signal becomes useful before it leaves the place.
  • Decentralized infrastructure. Computation and storage that does not require a hyperscaler in another jurisdiction. We study how to build research infrastructure that is genuinely distributed — across nodes, across institutions, across communities — without sacrificing reliability or open science.
  • Data sovereignty. A community's observations should stay under its control. Decentralized and edge architectures are how that becomes a default, not a promise. Sovereignty does not mean a wall — it means the right to share, withhold, attribute, and revoke on terms the community sets.
  • Measuring the footprint. Honest energy and carbon accounting for the AI systems we build, including the parts most people skip — training, fine-tuning, retrieval, inference, and the data movement around all of it.

How we approach it

Efficiency and sovereignty are not luxuries layered on after a system works. They are design decisions made at the start. We build with them assumed, and we publish the trade-offs so others can challenge our choices.

Read next