no formula?
learn it from the data.
Blackbox optimization, also called surrogate optimization, finds the most promising next decision from past data. kint learns how the system responds to past decisions, then searches that learned model. Each answer states how sure the model is.
your history already records how it behaves.
Some systems are too tangled to write down. Weather, vessel mix and crew habits interact on a quay. A furnace responds to twenty settings at once.
terminal throughput
Which yard and crane settings moved the most boxes per hour, learned from operational logs.
process settings
Temperatures, speeds and mixes that give high yield with few rejects, learned from production data.
flexible load
How demand and assets respond to price signals, learned from metering and market history.
from history to the next decision in four steps.
Your team decides what may be changed and what counts as a good outcome. kint does the modelling and the search.
bring the data
Past decisions and their outcomes from a database, an API or a spreadsheet. Every upload is versioned.
agree the structure
Which columns you control, which you cannot, which outcome to improve, which limits must hold. kint's decision agent proposes, your experts confirm.
train the surrogates
An ensemble of model types learns each outcome. kint checks the fit on data the models have not seen before anything is optimized.
search for better options
The optimizer searches the learned model within the stated limits, for one goal or several at once, and returns ranked candidates.
it says when it is unsure.
- uncertainty rangePredicted outcomes come with an uncertainty interval, not a single number.
- outside your dataA learned model only knows the history it was trained on. Candidates outside that range need a closer look before anyone acts.
- what drives itWhich inputs move the outcome, and in which direction, per candidate.
- several goalsCost against quality against emissions: the set of best compromises on the learned model.
- gets sharperNew outcomes flow back in. Retrain, and the next answer rests on more evidence.
blackbox or whitebox?
Both run on the same platform and return the same answer format. The difference is what you start from, and what the result can promise.
| Whitebox | Blackbox | |
|---|---|---|
| You start from | WhiteboxRules you can write down | BlackboxPast decisions and outcomes |
| Formula needed | WhiteboxYes, the agent drafts it with you | BlackboxNo |
| The result promises | WhiteboxBound and gap stated; optimal within your rules at 0.0 % | BlackboxBest found on the learned model, with an uncertainty range |
| It explains with | WhiteboxBinding limit and sensitivity | BlackboxInput influence and trade-offs |
| It improves when | WhiteboxYou refine the rules | BlackboxNew outcomes arrive |
| Best for | WhiteboxKnown structure, hard limits | BlackboxUnknown relationships, rich history |
bring your history. see what it knows.
Bring a sample of past decisions and outcomes. We train a first surrogate with your team and show you what it finds, and where it is unsure.
- Models
- Ensembles of gradient-boosted trees and Gaussian processes
- Answer
- Ranked candidates with uncertainty range
- Contact
- hello@kint.pro