Mynd Labs ResearchMynd Labs Research
Research for every field we work in.
Mynd Labs Research is the research company of Mynd Labs. An AI lab at the core, and open programs in fifteen fields: healthcare to agriculture, finance to energy.
Who we are
We begin with the question a field cannot yet answer, build the instrument that could measure it, and publish what we learn. The AI Lab is the core. Every field Mynd Labs enters gets its own program.
The AI Lab
The layer under everything.
Every Mynd Labs field runs on the same layer: models, agents, memory and tools. The AI Lab is where that layer is questioned from first principles, before any field depends on it.
- 01
Agent runtime
How an agent run becomes a thing you can pause, inspect, replay and compare, instead of a black box that either worked or did not.
- 02
Memory and context
What an assistant should remember, forget and refuse to carry across tasks, and how to prove it did what it said.
- 03
Evaluation
Tests that reflect real work. We want measures a field expert would sign off on, not only leaderboard numbers.
- 04
Safety and control
Permission, injection and failure behaviour in long, chained workflows, studied where they actually break.
- 05
Efficient intelligence
Doing more with smaller, cheaper and more local systems, so capability is not reserved for whoever can afford the largest cluster.
Programs
Fifteen fields, one question each.
Each program starts from one plain question. Pick a field.
Open questionHealthcare
How do you give clinicians more time with patients without taking judgement away from them?
Open program →
Open questionFinancial Services
Can every automated financial decision carry its own audit trail by construction?
Open program →
Open questionLegal
What does it mean for a system to read a contract the way a careful lawyer does?
Open program →
Open questionEducation
How should a learning system respond to the student in front of it, not the average one?
Open program →
Open questionGovernment
What does transparent automation look like when the user cannot opt out?
Open program →
Open questionRetail & E-commerce
How does a shopping system stay useful to the buyer and honest about the seller?
Open program →
Open questionManufacturing
Can a factory floor explain itself to the people who run it?
Open program →
Open questionMedia & Publishing
How do you scale creative work without losing provenance?
Open program →
Open questionTelecom
What does a self-explaining network look like?
Open program →
Open questionInsurance
How do you make a claim decision both fast and fair?
Open program →
Open questionReal Estate
Can the paperwork of a property be as clear as the property itself?
Open program →
Open questionLogistics
How should a supply chain behave when the plan breaks?
Open program →
Open questionEnergy & Utilities
How do you balance a grid that is getting more variable every year?
Open program →
Open questionNonprofits
How can small teams get big-organisation capability without big-organisation budgets?
Open program →
Open questionAgriculture
What can a smallholder know that used to take a lifetime or a laboratory?
Open program →How we work
Question. Instrument. Evidence. Limits.
- 01
Start from the question
A research program begins with something a field cannot currently answer, written down plainly. If we cannot state the question, we are not ready to build.
- 02
Build the instrument
We make the smallest tool that can measure the thing: a test, a trace, a dataset, a simulator. The instrument comes before the product.
- 03
Test against real work
Results count when someone who does the job would trust them. Expert review is part of the method, not a final step.
- 04
Say what we do not know
Every finding ships with its limits. We would rather publish an honest partial answer than a polished claim.