In development—
In development
Applied AI
The next product is being formulated. It applies language models to a workflow where the current state of the art is a spreadsheet and a lot of manual reading.
AI applicationWeb
Independent technology studio & research lab
Building intelligent productsthrough mathematics, artificialintelligence and data science.
Lluyot Labs is an independent studio that designs, models and ships software. We work across applied mathematics, machine learning and engineering — from the formulation of a problem to the product that finally solves it.
- Artificial Intelligence
- Machine Learning
- Mathematical Modelling
- Optimization
- Software Engineering
- Data Science
- AI systems
- Mobile applications
- Web platforms
- Mathematical software
- Data science tools
- Research projects
- Open-source libraries
- APIs
- Developer tools
- Simulation & modelling
- Decision systems
01How we work
The distance between a model
and a product.
Most of the value in applied mathematics is lost in transit. A correct result that never reaches a decision has produced nothing. The studio is organised entirely around closing that gap.
problemmodelanswerdecision
01
Formulate
We start by writing the problem down precisely: variables, constraints, objective, and what would count as being wrong. Most failed projects fail here, quietly, months before anyone notices.
02
Model & solve
Then we choose the smallest machinery that can answer it — a closed-form model, an optimizer, a learned model, or occasionally a well-chosen heuristic. Complexity is a cost, never a credential.
03
Ship & measure
A result that never leaves a notebook has produced nothing. We build the product around the model, instrument it, and keep the assumptions visible to the people relying on them.
02Technology
Six disciplines, used together.
We do not specialise in a single technique. Real problems rarely respect the boundary between statistics, optimization and engineering, and the interesting work usually sits exactly on it.
01
Artificial Intelligence
Systems that reason over language, images and structured data — designed around what the model is actually reliable at.
- LLM applications
- Retrieval systems
- Agents & tool use
- Evaluation harnesses
02
Machine Learning
Models trained on real data, validated honestly, and shipped with the uncertainty they actually carry.
- Forecasting
- Classification
- Anomaly detection
- Personalisation
03
Mathematical Modelling
Turning a messy real-world problem into a formulation precise enough to compute with — and simple enough to explain.
- Physical & biological models
- Simulation
- Numerical methods
- Sensitivity analysis
04
Optimization
Finding the best decision under real constraints: time, cost, risk, capacity, physics.
- Scheduling & routing
- Resource allocation
- Pricing
- Operations research
05
Software Engineering
The part that decides whether any of the above ever reaches a user. Built to be maintained, not demoed.
- Web & mobile apps
- APIs & services
- Developer tools
- Infrastructure
06
Data Science
Statistics before dashboards. Understanding what the data can and cannot support before anyone acts on it.
- Experiment design
- Causal inference
- Analytics pipelines
- Data visualisation
03Products
What the studio has shipped, and what comes next.
Coming soon—
Planned
Developer tooling
A tool for the parts of modelling work that are still done by hand. Expected to ship first as an open-source library, then as a hosted service.
Developer tool · Open sourceLibraryAPI
Coming soon—
Planned
Decision systems
An optimization product for operational planning — the class of problem where a two per cent better schedule is worth more than any interface.
Optimization · Web platformWebAPI
04Research
Mathematics is the part that decides whether it works.
Optimization, statistical inference and machine learning are not three separate departments here. They are three ways of answering the same question — what should be done, given what is known — and the choice between them is a modelling decision, not a technology preference.
Level sets of fdescent along −∇f
05Trajectory
A young studio, stated plainly.
Lluyot Labs is early. Rather than dress that up, here is exactly where it stands — what has shipped, what is being worked on now, and what is intent rather than fact.
- 2026Shipped
Lluyot Labs is founded
An independent studio built on a simple premise: the distance between a good mathematical model and a product someone actually uses is where most of the value — and almost all of the difficulty — lives.
- 2026Shipped
First engine: photobiological dose modelling
The studio's first production model. Erythemal and vitamin D accumulation integrated over real hourly irradiance, calibrated against published dose–response thresholds and validated for monotonicity before any interface was drawn.
- 2026Shipped
Broncea ships on Google Play
The first public product. Six languages, no accounts, no advertising, no server holding user data — and a screen inside the app that shows the evidence behind every number it reports.
- NowIn progress
Second product in formulation
An applied AI product, currently at the stage the studio considers most important and least visible: writing the problem down precisely enough to know what a correct answer would look like.
- NextPlanned
Open-source foundations
Extracting the reusable parts of the modelling work into libraries, released publicly. Tools that are used by other engineers get corrected by other engineers.
- NextPlanned
Research collaborations
Working with academic groups and companies on problems where the modelling is genuinely open. Selective by necessity, and only where the studio can contribute something specific.
06Vision
Every meaningful problem eventually becomes a mathematical one. Our work is to get it there — and then to build the thing that makes the answer usable.
Where we are going
A portfolio of products, each built on a model we understand completely, in domains that are unrelated to one another. The through-line is the method, not the market. A studio that can do this well in one field can do it in the next one, and that is the only durable advantage we are trying to build.
What will not change
The numbers a product reports are either defensible or they are not shown. The assumptions behind them stay visible to the person relying on them. And what we say about our own work stays accurate, including the parts that are less impressive than they could be made to sound.
Work with us
The studio is small and selective.
That means we answer email personally, and that we say no to work we cannot do properly. If you have a problem that needs modelling rather than staffing, describe it and we will tell you honestly whether we are the right people.