Aesthetical Quality Control for Luxury Manufacturing

Subpixel Image Alignment

Developed and benchmarked advanced image alignment algorithms to register high-resolution captures of watch components with subpixel precision.

→ Sub-pixel accuracy on production imagery

Large-Scale Image Manipulation

Engineered pipelines for manipulating very high-resolution images (multi-gigapixel composites) that exceed GPU and system memory limits. Implemented tiling strategies with overlap management, lazy-loading architectures, and streaming processing that enable real-time analysis of massive surface captures without resolution compromise.

→ Multi-gigapixel processing on standard hardware

Image Processing & Feature Engineering

Researched and implemented specialized dimensionality reduction techniques to transform high-dimensional image data into discriminative, compact feature representations. Explored manifold learning approaches tailored to the specific structure of surface inspection data, enabling more effective downstream anomaly detection.

ML & Deep Learning Anomaly Detection

Built a comprehensive anomaly detection pipeline that combines classical machine learning approaches with deep learning architectures for identifying aesthetical defects. Each model family brings complementary strengths statistical robustness from classical methods and pattern sensitivity from deep networks enabling high recall with controlled false-positive rates.

→ Production-grade detection accuracy

Depth Estimation & 3D Reconstruction

Explored 3D depth estimation and surface-reconstruction methods to recover physical geometry from image captures covering depth, normal, texture, and shape reconstruction through both classical and neural approaches. The goal: enable inspection that accounts for surface geometry curvature, micro-relief, and local deviations rather than 2D appearance alone.

Software Architecture & Data Pipelines

Designed the overall software architecture for the quality control system from image acquisition through processing stages to result storage and visualization. Built robust data pipelines that orchestrate multi-stage workflows, handle error recovery, and maintain data lineage across the full inspection chain. The architecture supports modular algorithm swapping and configuration-driven pipeline composition.

Vector DB Search & Similarity Retrieval

Integrated vector database search for rapid similarity retrieval across high-dimensional feature embeddings. This enables defect matching against known reference databases, nearest-neighbor classification of anomaly types, and fast lookup of similar historical cases critical for both automated decision-making and operator-assisted review workflows.

Memory & Speed Optimization

Optimized the entire processing pipeline for production-grade performance implementing concurrent processing patterns, efficient GPU memory management, batched inference, and parallel computation to meet throughput requirements. Profiled and eliminated bottlenecks across the full stack, from I/O to GPU compute, achieving real-time processing on standard industrial hardware.

→ Real-time processing throughput

MLOps & Deployment

Managed the full MLOps lifecycle from experiment tracking through model packaging to production deployment. Built containerized inference services delivered as POCs and pilots, with monitoring, logging, and automated health checks. Supported iterative model improvement cycles with versioned deployments and A/B testing capabilities.

→ Deployed in production

UI/UX Development & Statistical Reporting

Designed and developed the complete user interface for the quality control application including real-time inspection visualization, operator workflows, result browsing, and statistical dashboards. Built reporting modules to track quality KPIs and process metrics over time.

→ Actionable reporting for operators