Industrial machine learning

Industrial machine learning,
without the project.

A break from 12-to-18-month AI deployments: your raw data, proprietary algorithms, an intuitive platform and your AI assistant, in a single product.

For the plants that will never have a data scientist. A CSV export is enough to start.

Reply within 24 h. No credit card, no time limit.

Four building blocks, one productfrom raw data to the answer
04 · Your AI assistant

Does a serious stop start within the next hour?

MCP connectionJemba, available from your assistant
03 · Intuitive platformA model re-run in one click
02 · Proprietary algorithms
Anomaly detectionOptimisation
01 · Your raw data
.csvOPC UAHistoriansReal-time APITEEPTRAK
AccessFree, with no time limit
DataFrom raw historical .csv to real-time API
AI assistantMCP connection to AI assistants
Proof of value

What the data from a single line has already shown.

Four results from proofs of value run with manufacturers, on their own data.

Aerospace & defence
€215k

of avoidable downtime identified on one line

Aerospace & defence
273h

of downtime anticipated, 45% of the line’s total
from the run/stop event log alone

Read the case →
Rubber & polymer
30%

less scrap on one line

Rubber & polymer
+9.5pts

of material yield on a bobbin line
977 columns screened down to 10 levers

Read the case →

Proofs of value, not large-scale deployments. The manufacturers are not named.

Our algorithms

Proprietary algorithms, from our own research.

Two algorithm families: anomaly detection and optimisation.

We also publish the machines it didn’t work on.

thresholdalert
Family 01

Anomaly detection

“Does a serious stop start within the next hour?”

An anomaly score and an adaptive threshold, machine by machine.
Learn more →
setpoint rangeyieldsetting
Family 02

Optimisation

“Where should the line run so the yield holds?”

The output is a setpoint range to hold. Not a notebook.
Learn more →
01

A team of PhDs. They design Jemba’s algorithms.

02

Our own research. Our algorithms are proprietary.

03

Reliable results. Across a wide range of situations and data types.

Platform

A scalable platform, and it is free.

Free access includes 2 anomaly detection projects and 1 optimisation project, with no time limit. You choose the target variable and re-run the model in one click.

2anomaly detection projects
1optimisation project
no time limit
Anomaly score across all lines, with the models currently in production
Anomaly score across all lines, with the models currently in production
Request my free access

Reply within 24 h. No credit card, no time limit.

Your data

Your data, whatever its level of maturity.

Your plant has been instrumented for a decade. The data died in storage. Jemba starts from there.

.csvRaw history

Your exports, as they are today.

OPC UAMachines and sensors

Connected over OPC UA.

HistoriansYour historians

The years of history you already keep.

APIReal-time API

Your data as a continuous stream.

And alsoTEEPTRAK

Your TEEPTRAK production data.

See the documentation →
AI assistants

Run Jemba from your assistant.

The assistant reasons and explains. Jemba computes, monitors continuously and guarantees the result.

01

Three ways to use Jemba. Through the platform, through a chat, or from your own AI assistant.

02

The assistant you have chosen. Jemba plugs into it through an MCP connection. You ask your shop-floor question the way you would say it.

03

Your data stays yours. A client’s data only serves that client.

Your AI assistantYour questions, in plain language
MCP connection
JembaYour data, our algorithms, the platform
Getting started

Three steps to your first result.

1Request your accessWe reply within 24 h.
2Upload your dataA raw historical .csv is enough to start.
3Read your first resultAn anomaly score or a setpoint range.
Request my free access

Reply within 24 h. No credit card, no time limit.

We use this information only to reply to you. No mailing list, no resale.