THELIO SOLUTIONS
EN / FRBook a Call
OUR WORK

AI applied to real operational problems.

A selection of projects spanning industrial computer vision, document intelligence, optimization, sensor analytics and scientific imaging. Client-identifying details are intentionally omitted where confidential.

Document Automation

Customer Order Intake Automation

Interpreting highly variable customer orders and mapping them to standardized SAP terminology.

ChallengeEach customer used different layouts, naming conventions and terminology, while internal processing required consistent SAP-ready information.
ApproachAn intelligent document-processing pipeline reads incoming orders, understands context and maps customer-specific terminology to internal company references.
Document AINLPSAPAutomation
Document Automation

Automated Invoice Processing

Extracting, structuring and validating invoice data to reduce repetitive manual processing.

ChallengeInvoice layouts and field locations vary considerably across suppliers, creating time-consuming manual data entry.
ApproachA document-understanding workflow extracts key fields, normalizes them and prepares them for downstream validation and business systems.
OCRDocument AIValidationWorkflow
Document Automation

Technical Drawing Part-Name Extraction

Automatically identifying and extracting part information from engineering drawings.

ChallengePart names were embedded in complex technical drawings where conventional text extraction was unreliable.
ApproachA layout-aware extraction approach combines document structure and text understanding to identify the relevant part information.
Technical DrawingsOCRExtraction
Document Automation

Intelligent Document Classification

Classifying technical documents by both content and visual layout.

ChallengeIncoming files included drawings, customer specifications and other document types with overlapping terminology.
ApproachA multimodal classification pipeline uses textual and layout information to route documents to the correct downstream workflow.
ClassificationLayoutDocument AI
Document Automation

Privacy-Preserving Technical Drawing Extraction

Detecting and isolating sensitive drawings before documents are sent to cloud services.

ChallengeTechnical drawings contained sensitive information that could not leave the protected environment, while other document content could benefit from cloud processing.
ApproachA local computer-vision stage detects and removes restricted drawing regions before the remaining content enters the cloud workflow.
PrivacyComputer VisionCloudDocument AI
Computer Vision

Microscopic Multi-Layer Coating Detection

Detecting and characterizing multiple coating layers in microscopic metal images.

ChallengeLayer boundaries in microscopic cross-sections can be subtle, irregular and time-consuming to assess manually.
ApproachA computer-vision workflow segments the coating structure and supports repeatable measurement of multiple microscopic layers.
MicroscopySegmentationIndustrial AI
Computer Vision

Scratch & Contamination Detection

Detecting scratches, dust and surface imperfections in microscopic coating imagery.

ChallengeSmall visual defects were difficult to inspect consistently at scale.
ApproachA vision model identifies and localizes defect patterns to support faster, more consistent inspection.
Defect DetectionMicroscopyQuality
Computer Vision

Coating Simulation Object Tracking

Tracking parts through a coating simulation to assess repeatability and surface exposure.

ChallengeThe team needed to verify repeatable movement and whether all sides of a component were exposed during simulated coating.
ApproachObject detection and tracking quantify the component trajectory and support coverage analysis across repeated runs.
Object TrackingSimulationComputer Vision
Computer Vision

3D Printing Powder-Bed Anomaly Detection

Detecting irregularities in additive-manufacturing powder-bed images during production.

ChallengePowder-bed anomalies can affect downstream part quality and are difficult to monitor manually layer by layer.
ApproachAn image-based anomaly-detection pipeline flags unusual patterns for inspection during the manufacturing process.
Additive ManufacturingAnomaly DetectionVision
Computer Vision

CT-Based Fatigue Estimation

Using CT scans of additively manufactured cylinders to estimate fatigue-related characteristics.

ChallengeInternal material characteristics visible in CT data needed to be connected to fatigue behavior.
ApproachImage-derived features and machine-learning methods are used to model fatigue-related outcomes from volumetric scan data.
CT3D ImagingMachine Learning
Machine Learning

Yarn Sensor Pattern Classification

Multi-class classification of sensor patterns generated during yarn production.

ChallengeComplex sensor signatures had to be mapped to distinct production patterns and operating states.
ApproachA supervised machine-learning pipeline learns discriminative patterns from multivariate sensor data and classifies production behavior.
Sensor DataClassificationManufacturing
Machine Learning

AI-Assisted Knee MRI Classification

Multi-label machine-learning analysis of knee MRI studies for abnormality classification.

ChallengeMRI studies contain multiple imaging series and can present several abnormalities simultaneously.
ApproachA deep-learning pipeline aggregates information across MRI series and produces multi-label predictions across several abnormality categories.
MRIDeep LearningMulti-label Classification
Machine Learning

Pickup Routing Optimization

Optimizing customer pickup routes to reduce unnecessary travel and improve planning.

ChallengePlanners needed an efficient way to sequence multiple customer stops while considering distance and operational constraints.
ApproachA route-optimization workflow scores candidate stops, builds efficient trip sequences and supports interactive planning.
OptimizationRoutingDecision Support