Java-native AI, proven in production.
From payment fraud to particle physics – teams run Deep Netts where their systems already live: inside the JVM. Four real deployments, each without adding a second stack.
Real-time fraud detection inside the transaction flow
- Problem
A payments platform needed to score every transaction for fraud in real time – but routing traffic through a separate Python scoring service added latency, infrastructure, and a second stack to operate.
- Solution
Deep learning models built with Deep Netts run directly inside the Java transaction pipeline. Scoring happens in-process, with no network hop and no additional runtime to deploy or monitor.
- Outcome
Fraud decisions now land in milliseconds, inside the same JVM that processes the payment – and the fraud team iterates on models with the tools the platform team already runs.
Forecasting cloud resources before the bill arrives
- Problem
An enterprise platform team was over-provisioning cloud capacity to stay safe, because usage forecasts were coarse and reactive – and the cost of that safety margin kept growing.
- Solution
Deep Netts models forecast resource demand from historical usage, running natively inside the team's JVM-based tooling. Predictions feed autoscaling and capacity planning directly, with no separate ML pipeline.
- Outcome
Provisioning now follows predicted demand instead of worst-case guesses, cutting spend while improving stability during load spikes.
Illustrative data. Hover or use the arrow keys to read each month.
| Month | Demand | Provisioned | Fixed before |
|---|---|---|---|
| Jan (actual) | 520 | 572 | 920 |
| Feb (actual) | 560 | 616 | 920 |
| Mar (actual) | 540 | 594 | 920 |
| Apr (actual) | 600 | 660 | 920 |
| May (actual) | 640 | 704 | 920 |
| Jun (actual) | 620 | 682 | 920 |
| Jul (actual) | 690 | 759 | 920 |
| Aug (actual) | 710 | 781 | 920 |
| Sep (forecast) | 730 | 803 | 920 |
| Oct (forecast) | 760 | 836 | 920 |
| Nov (forecast) | 800 | 880 | 920 |
| Dec (forecast) | 770 | 847 | 920 |
Predicting particle trajectories with neural networks
- Problem
Reconstructing particle trajectories from detector data with traditional statistical methods was computationally expensive, limiting how fast experiment data could be processed.
- Solution
Jefferson Lab replaced the statistical approach with Deep Netts neural networks trained on detector data – running in the lab's existing Java-based analysis infrastructure.
- Outcome
Trajectory prediction became significantly more computationally efficient, letting the lab process experiment data faster on the same hardware.
- 138
- Hits recorded
- –
- Kept (on tracks)
- –
- Rejected as noise
- –
- Tracks found
Illustrative hits. CLAS12: 6 sectors × 3 drift-chamber regions. Hover a track to isolate it.
Machine learning for advanced materials research
- Problem
A materials research group needed ML models for property prediction and data processing, but the surrounding tooling was Java – adding a Python stack meant new dependencies and maintenance for a research team.
- Solution
The group builds and trains models with Deep Netts directly in Java, keeping the whole workflow – data processing, training, prediction – inside one ecosystem.
- Outcome
Faster data processing and property prediction, with no new language stacks or dependencies added to the research environment.
42 of 48 predictions within ±0.05 of the reference (shaded band).
Illustrative data, normalised property values. Hover a point for details.
One stack instead of two.
In every case the model moved out of a separate service and into the Java application that already handles the data.
- – Network hop on every prediction
- – Second language, second deployment, second on-call
- – Data leaves the application
- + Predictions in-process, in milliseconds
- + One language and one toolchain for the team
- + Data stays inside your environment
Have a workload like these?
Tell us about your system – we'll show you how Deep Netts fits inside it.