ML TWO/Service Catalogue/ML2Pulse
Service 07

ML2Pulse

Predict, anticipate, act

Machine Learning systems built around two concrete objectives: Anomaly Detection and Predictive Maintenance, spotting anomalous behavior and anticipating failures before they occur. Around this core, applied research expertise tackles high-complexity industrial problems with academic rigor, through an end-to-end MLOps cycle from training to retraining in production.

Let's TalkAreas of intervention
Anomaly DetectionPredictive MaintenanceApplied researchEnd-to-end MLOpsMEPA-registered
Service Sheet

Predict failures before they happen

Machine Learning for Anomaly Detection and Predictive Maintenance, with applied research for complex problems.

ML2Pulse

Predict, anticipate, act

Machine Learning systems focused mainly on Anomaly Detection and Predictive Maintenance: spotting anomalous behavior and anticipating failures before they occur. Around this core, applied research expertise tackles high-complexity industrial problems with academic rigor, with an end-to-end MLOps cycle from training to retraining in production.

Industrial ML Predictive and prescriptive models: anomaly detection on time series (IoT, sensors, SCADA), predictive quality, multivariate forecasting and process optimization with reinforcement learning.
Computer Vision Automatic image and video analysis: visual inspection for quality control, semantic segmentation, medical imaging, satellite and drone monitoring, 3D reconstruction.
Optimization Tailored algorithms for combinatorial problems: production scheduling, routing and logistics, multi-objective resource allocation, metaheuristics and Monte Carlo simulation.
NLP & Knowledge Domain-specific natural language: fine-tuning LLMs for vertical domains, information extraction, knowledge graphs, advanced RAG with traceability for regulated contexts.
AI for Scientific Computing ML applied to numerical simulations: Physics-Informed Neural Networks, surrogate models for CFD/FEM, predictive digital twins, causal inference.
AI Safety & Explainability Reliability and interpretability: explainability (SHAP, LIME), adversarial robustness, fairness audits, uncertainty quantification and model drift monitoring.
Financial Fraud Detection Fraud detection in banking and finance based on Graph Neural Networks, the state of the art in the field: models on transactions and payments, real-time risk scoring, detection of fraud rings and anomalous patterns, and AML (anti-money-laundering) support with fewer false positives. Our team has produced improvements to GNN algorithms validated through international scientific publications.

Applied research: doctoral-level expertise supporting the most complex use cases, where standard libraries fall short — from scientific prototyping to industrial deployment.

Anomaly Detection & Predictive Maintenance 07
Areas of Intervention

Areas of expertise

Each area can be engaged on its own or as an integrated path, scaled to your real needs.

01
Industry

Industrial ML

The predictive and prescriptive core: models that spot anomalies in process data and anticipate failures.

  • Anomaly detection on time series from IoT, sensors and SCADA
  • Predictive quality and multivariate forecasting
  • Process optimization with reinforcement learning

What you getFailures anticipated and process quality under control.

02
Vision

Computer Vision

Automatic image and video analysis for inspection and monitoring.

  • Visual inspection for quality control and semantic segmentation
  • Medical imaging and satellite and drone monitoring
  • 3D reconstruction

What you getAutomatic eyes on quality and territory.

03
Algorithms

Optimization

Tailored algorithms for the hardest combinatorial problems.

  • Production scheduling, routing and logistics
  • Multi-objective resource allocation
  • Metaheuristics and Monte Carlo simulation

What you getOptimal decisions under real constraints.

04
Language

NLP & Knowledge

Domain-specific natural language to extract and organize knowledge.

  • Fine-tuning LLMs for vertical domains
  • Information extraction and knowledge graphs
  • Advanced RAG with traceability for regulated contexts

What you getKnowledge that is queryable and traceable.

05
Research

Fast Simulations and Models You Can Trust

Engineering simulations — fluids, structures, heat — can take hours of compute. We train models that reproduce their results in seconds, and we verify that every prediction can be explained before it goes into production.

  • Simulations reproduced in real time by models that respect the physics of the phenomenon (surrogate models, Physics-Informed Neural Networks)
  • Digital twins that predict how plants and processes behave as conditions change, telling real causes apart from coincidences
  • For every prediction we show the reason behind the result and its margin of uncertainty, not just the final number
  • Regular checks for bias, manipulability and degradation of the model over time

What you getAnswers in seconds instead of hours of compute, and numbers you can explain to a client or a certification body.

06
Finance

Financial Fraud Detection

Fraud detection in banking and finance built on the state of the art: Graph Neural Networks (GNN), which analyse the network of relationships between accounts, devices and beneficiaries rather than each transaction in isolation. This is the approach that catches organised fraud, where no single movement looks suspicious on its own.

  • Models on transactions and payments with real-time risk scoring
  • GNNs to surface fraud rings and anomalous patterns that traditional models miss
  • AML (anti-money-laundering) support with fewer false positives to work through
  • Improvements to GNN algorithms developed by our research team and validated through peer-reviewed international scientific publications

What you getFraud intercepted even when it hides inside a network of accounts, and fewer false positives to clear. References to the published work can be shared in a follow-up conversation.

How We Work

A Four-Step Path

Every project starts from where you really are: a no-commitment initial assessment to define priorities and goals.

01
Scientific framing

We frame the industrial problem and the available data with academic rigor.

02
Prototyping

We develop the model from scientific prototyping, even where standard libraries fall short.

03
Deployment

We bring the model into production with an end-to-end MLOps cycle.

04
Retraining

We monitor model drift and manage continuous retraining in production.

Let's Talk

Let's Talk, No Commitment

Tell us your need: you'll get a clear picture of the priorities, even if the path continues without us.

Contact ML TWO

Or write directly to info@mltwo.it