04 / TECHNICAL AREA

AI

Measurable intelligence, in the real world

Artificial intelligence for data, images, audio and signals, from software to devices. I can support your team on datasets, models and integration, adapting to existing libraries, pipelines and infrastructure. Objectives and verification follow the problem and real operating conditions.

Platforms and tools follow your context. Some examples:

Python / NumPy / C++ / TinyML / Computer vision / Edge AI

Skills & technologies

AI

Custom AI · models · inference

Custom AI solutions
Goals · data · workflowsSolutions for specific use cases
Machine learning
Classification · regressionModels and real-use-case metrics
Computer vision
Images · detection · inspectionAcquisition and visual analysis
Audio & signals
Classification · analysisAudio, sensors and time series
Edge AI & TinyML
MCUs · SoCs · acceleratorsOn-device inference
Datasets & evaluation
Data · training · validationModel comparison and verification

The technologies shown are examples.

Engineering support within your team, with shared tools and working methods.

Open graphic ↗

Machine learning and computer vision with a verifiable objective

I can join your team’s work on data, models and software or take on a jointly defined experiment. I adapt to existing libraries, computing environments, repositories, pipelines and evaluation criteria. The solution follows the client’s problem and infrastructure, from backends to devices.

A machine learning project starts by defining the task: what to recognize, which errors matter and which data will verify the result. Before selecting a model, we assess example availability and representativeness, operating conditions and acquisition methods. Metrics need to address the use case and remain separate from training data.

Industrial computer vision includes images, lighting, viewpoint and inspection software interfaces. Classification and anomaly detection need to be assessed alongside acquisition robustness. Silicon LogiX demonstrators also explore synthetic images and model-version comparisons, with traceable experiments and results.

For audio and signals, acquisition and data preparation form part of the solution. A classifier running on a device needs inputs consistent with those evaluated during development. Integration connects the model, signal processing and application logic while keeping outcomes and fault conditions observable.

Edge AI & TinyML: inference within device constraints

Running inference on microcontrollers or edge platforms requires comparing accuracy, memory, power and execution time. TinyML, available runtimes and target accelerators are evaluated in the product context. The choice between local processing, a backend or a hybrid architecture also depends on connectivity, updates and data handling.

The result is more than a model file: scope can include data preparation, inference software, target testing and documentation of operating conditions. Projects and articles in this area explore audio ML, computer vision, NPUs and on-device inference through documented cases and engineering decisions.