Oscar Efrén Narváez
Co-founder · Director and president
Leads strategy, product development and market research.
We are building AI agents that keep watch over an industrial plant, inside limits its owner signs. A fault at two in the morning becomes one tap on your phone, not a drive to the plant.
Pre-seed, from Managua, Nicaragua. A working prototype runs on a digital twin of a reverse-osmosis water plant. We are preparing the first pilot on a real plant.
Normalized flow is down 13.9% since the last clean and the permeate tank is at 58%.
Start a chemical clean (8 hours). It needs your approval: it pauses water delivery to the mine.
Try it. Synthetic data from our digital twin.
People who own a plant tell us the same thing. They want more time for family, for travel, for the next business. What stops them is the fear of what happens when they are not there.
Something drifts, an alarm goes off, and the decision waits until the owner picks up the phone.
The data is already there, in the PLC and the SCADA. What happens with it depends on who is on shift.
So the owner stays close. Lux Cognition exists to change that: the plant is watched, and the owner decides only what truly needs them.
Each agent looks after one part of the plant: process, supplies, maintenance, instruments, quality. None of them can step outside what you sign.
The agents read what the plant already measures and check that each instrument can be trusted before using it.
Small, low-risk adjustments happen on their own, inside a signed range, step and minimum interval.
Anything bigger waits for a person. You see what, why and what it costs, and you approve or decline.
Every proposal, action and decision goes into a log that shows who, when and exactly what was written.
Between the agents and the plant there is a guardian. It is the only part allowed to write, and it checks every request against the limits you signed.
Emergency stops and safety systems are off limits. Always.
Which valves and pumps, how far, how fast and how often. Outside that, the answer is no.
If an instrument is frozen, out of range or flagged as bad, anything that depends on it is blocked.
Figures come from process calculations. The language model only explains them and its text is rejected if it changes one.
The models can run on a computer inside the plant. Client data is not sent to free or public AI services.
Before it touches anything, Lux only records what it would have done, so you can compare it with your operators.
We test the guardian with adversarial cases, and each one that ever worked is now an automatic test.
We would rather show the real stage than promise more than we have tested.
On a digital twin, with synthetic data.
The road to version 1.0.
To get from prototype to first plant.
A reverse-osmosis plant that delivers 8 cubic metres of water an hour to the mine's recovery process. If the water stops, the process stops. That is the plant our digital twin reproduces.
Lux Cognition is a Central American company based in Managua. We put process engineering in the field together with artificial intelligence.
Co-founder · Director and president
Leads strategy, product development and market research.
Co-founder · Engineer
Leads OT/IT integration, quality and key accounts.
If you own or run a plant and would like your time back, or you want to follow what we are building, write to us.