Computer Vision · Medicine
Medical image segmentation
Context: Segmentation of target structures on MRI data, classification and visual quality metrics.
Case studies
Machine learning and data analysis
Models built for a specific task and dataset: computer vision, audio analytics, clustering of large datasets, and generative geometry for manufacturing.
8 projects in this category · ML, LLM, RAG and analytics
Computer vision
Task: Automate quality control.
Solution: CV models and real-time visual analytics.
Result: Metrics: defects down ~27%, inspection speed up ~38%, defect recall +14 pp.
Computer Vision · Product
Image detection and analysis
Context: Composition recognition from photos with a detailed report — a product interface over CV models.
Audio analytics: voice spoofing detection
Task: Reliably distinguish genuine from spoofed audio.
Solution: A combination of STFT/MEL/MFCC/FFT features and a model ensemble (CNN, Logistic Regression, Gradient Boosting).
Result: Accuracy/F1 up to 0.9992, ROC-AUC up to 1.0000, EER about 0.31%, stable class separation in UMAP.
Audio ML
Audio Anti-Spoofing — metrics
Context: Model comparison by Accuracy, F1-Score and ROC-AUC.
Audio ML
Audio Anti-Spoofing — features
Context: Bonafide/spoof classification on STFT, MEL and MFCC features.
Data Science
Audio Embeddings — UMAP 3D
Context: Class clustering in 3D latent space.
Data Science
Audio Embeddings — UMAP 2D
Context: Audio class separation by feature type (raw/mel/stft/mfcc/fft).
Energy grid clustering
Task: Optimise service routes for 298K Arizona power grid assets.
Solution: K-Means + HDBSCAN, 4 regions, CQ priority scoring and a 4-level dispatch system.
Result: Metrics: incident response time down ~34%, SLA coverage 89% → 97%, operating costs down ~21%.
Data Science · Geoanalytics
Power grid clustering, 298K points
Context: K-Means + HDBSCAN over Arizona power grid users: 4 regions, CQ scoring, 4 service levels.
AI CAD for 3D printing
Task: Shorten the geometry preparation cycle for additive manufacturing.
Solution: Quadric approximation, surface parameterisation and CAD representation validation.
Result: Metrics: manual CAD edits down ~34%, time to print down ~41%, first-pass yield +16 pp.
R&D · Additive manufacturing
AI CAD for 3D printing
Context: Geometry approximation and CAD model preparation for additive manufacturing.
Discuss a project
Describe the task, the data and the timeline. We will propose a format and an outline plan. Reply within one business day, NDA on request.
Discuss a project