Custom models

Want a custom model?

Reflux trains models on your process data: converged cases, plant history and lab results. Predict what a case will do before it runs, and keep the model where your data lives.

Tell us what it should predict How an engagement runs

Scoped per engagement. Your data is never used to train anything for anyone else.

Trained on: Your converged cases, plant historian data, lab results
Predicts: Convergence and failure modes, yields and specs, scale-up behaviour between rigs, whatever you scope
Called from: The same Reflux agent, inside the simulator workflow
Runs in: Your environment, your private cloud, or Reflux-hosted
Belongs to: You

What it can predict

A general model knows chemistry. A model trained on your process knows your process.

Case outcomes before the run: will it converge, where it will fail, which spec will bind.

Surrogates for slow models so a sweep that takes an afternoon in the simulator takes seconds, with the simulator kept for the cases that matter.

Scale-up behaviour learned from your own history of lab, pilot and plant results, so the next candidate is ranked on evidence.

How an engagement runs

Scope
One target to predict, the data you have, and where the model is allowed to run. A call, then a one-page scope.
Train
Inside your environment where the data cannot leave it. Reflux-hosted where it can. Either way, your data trains your model and nothing else.
Deploy
Behind the same Reflux agent, so the model is called from the simulator workflow, not from a notebook someone has to remember.
Iterate
Every run adds data. The model improves on your process, and it stays yours.

ScopeConvergence risk, T-series columns

Target: probability a proposed case converges, with the likely failing stage.
Data: 1,400 historical cases with run logs, your historian for feed drift.
Runs in: your private cloud. Nothing leaves it.

Called from Reflux before each run

Tell us what you want it to predict

Three lines is enough. What the model should predict, roughly what data you have, and where it would be allowed to run. A person reads every one of these.

1
The target
The one number or outcome that would change a decision if you knew it earlier.
2
The data
Simulator runs, historian tags, lab notebooks, batch records. Messy is fine.
3
The boundary
Whether the data can leave your environment, and whether the model has to live inside it.

Work email

Name Company or lab

What should the model predict?

What data do you have? Where must it run?

No sequence, no drip. You get a reply from a person, and your data is never used to train anything for anyone else.

Questions people ask

Do you train on our data for other customers?

No. Your data trains your model, the model belongs to you, and nothing about it is reused for anyone else. Reflux itself is not trained on customer cases.

What do you need from us?

One target worth predicting and the data you already have, in whatever state it is in. Simulator run logs, historian exports and lab records are the usual starting points.

How long does it take?

Scoping is a call and a one-page document. A first model on existing data is weeks rather than quarters, and how many weeks depends almost entirely on the data.

What does it cost?

Scoped per engagement, against the target and the data. The scope document carries the number before any work starts.

Can it run without anything leaving our network?

Yes. Training and inference can both run inside your environment. Where the data is allowed to leave, a private cloud or a Reflux-hosted model is simpler and cheaper, and the choice is yours.