In development

One pipeline, from raw data to a decision you can defend.

The Sondera platform moves inspection data from ingestion through physics-grounded signal processing and detection models to traceable, audit-ready reports. Here is how it works.

How it works

Four stages from signal to decision.

Each stage produces a defined output that feeds the next, so the path from a raw scan to a signed-off report is explicit end to end.

  1. 01

    Ingest

    Upload scans in their native formats. Datasets are organized by part, campaign, and modality, and versioned so every later result traces back to an exact input.

    OutputVersioned dataset

  2. 02

    Signal processing

    Denoising, deconvolution, alignment, and feature extraction tuned to the physics of the modality. Raw measurements become clean, comparable, decision-ready data.

    OutputConditioned data

  3. 03

    Detection and characterization

    Physics-informed models locate indications, estimate size and depth, and classify defect types. Every call carries a confidence measure.

    OutputDefect map + confidence

  4. 04

    Decision and reporting

    Results map to your acceptance criteria and produce a traceable, audit-ready report that links each call back to the raw data and the model version behind it.

    OutputSigned-off report

Architecture

One core, every modality.

Sondera is built in layers. Modality adapters normalize incoming data so the signal-processing core, the detection models, and the reporting layer never have to care which instrument produced it. Terahertz runs on this core today; ultrasonic and micro-CT reuse the same pipeline as they come online.

A data and audit store sits under all of it, versioning inputs, results, and the model that produced them.

  1. Modality adapters

    Bring THz, ultrasonic, or micro-CT data into a common representation.

  2. Signal-processing core

    Shared, physics-grounded conditioning and feature extraction.

  3. Detection models

    Physics-informed ML for localization, sizing, and classification.

  4. Decision and reporting

    Acceptance criteria applied, reports generated.

  5. Data and audit store

    Versioned inputs, results, and model versions, kept for audit.

What it produces

Every run leaves an auditable trail.

Defect maps

Spatial maps of detected indications, registered to the part and comparable across campaigns.

Characterization with confidence

Size, depth, and type per indication, each reported with a confidence measure rather than a bare label.

Audit-ready reports

Structured reports that state the acceptance criterion applied and link every call to its evidence.

Versioned datasets

Inputs, results, and the model version that produced them, retained so any decision can be reproduced.

Traceable by default

Every decision links back through a fixed chain, so an audit is a lookup rather than an investigation.

Raw dataConditioned dataModel versionAcceptance criterionSign-off

The platform is being built with pilot partners.

We are onboarding a small group working with composite materials and terahertz data. Pilot partners get first access and shape the roadmap.