Edge & low-bandwidth AI
Compact models that run on low-cost phones and field devices, sync when connectivity returns and still give useful answers when it does not.
The Lab
Research with a deadline and a user in mind.
The alba.one lab is where we investigate questions before our clients have to bet on them. Each line of research is tied to a real industry problem, and each ends with evidence: a result, a prototype or an honest "not yet".
Research areas
What we are working on
Trustworthy & accountable AI
Evaluation methods, model cards and decision logs that make automated decisions explainable to customers, auditors and regulators.
Verifiable data & provenance
Linking physical goods, sensors and documents to on-chain records so claims about origin, quality and impact can be checked.
Tokenised real-world assets
Legal, economic and technical designs for representing land, harvests, receivables and carbon credits as digital assets.
Privacy-preserving computation
Zero-knowledge proofs, federated learning and secure data collaboration, so organisations can share insight without sharing raw data.
Agentic systems
LLM agents that operate inside business workflows, with bounded permissions, human checkpoints and full audit trails.
How the lab works
Rigour without the ivory tower
Principles
- Start from a decision someone needs to make.
- Agree what success looks like before the first experiment.
- Test on real, messy data as early as possible.
- Publish what we learn, including what did not work.
- Design for the constraints of the market, not the lab.
Toolkit
- Python, PyTorch, scikit-learn, ONNX for edge deployment
- Open and hosted LLMs, retrieval and evaluation frameworks
- Solidity, EVM chains, Hyperledger, zero-knowledge tooling
- PostgreSQL, streaming pipelines, geospatial and satellite data
- Docker, Kubernetes, CI/CD and observability
Method
From question to evidence
Frame
Turn a broad ambition into a sharp, testable research question.
Experiment
Run small, fast experiments against baselines and real data.
Prototype
Put the best approach in front of real users and measure it.
Recommend
A clear decision memo: scale, adjust or stop — with the numbers behind it.
Research partners welcome
We collaborate with companies, universities and funders on applied research programmes. Let us know what you are exploring.