Research
A record of our technical methodology. Algorithms, mathematical formulations, machine learning approaches, and reviews of industry literature.
- Architectural design of an Agent-Assisted Product Discovery Pipeline A customer submits a need in natural language. The system finds, vets, prices, and lists the product. What does the AI layer actually look like — is it a single model, an ensemble, a fine-tuned LLM? A design-first analysis of each sub-problem and the approach it warrants.
- Item2Vec for Sparse Purchase Histories: Embedding Products in Implicit Feedback Regimes How we adapted Word2Vec's skip-gram architecture to learn product embeddings from sparse, implicit purchase signals — without collaborative filtering at scale.
- ResNet-18 for Image Classification: Residual Learning on MNIST and CIFAR-10 How we implemented a ResNet-18 classification system with dual CLI/GUI interfaces, and what the residual connection formulation buys you over plain convolutional stacks.
- Semantic Segmentation of Land Cover from Multispectral Imagery: A CNN Approach Our methodology for training a convolutional segmentation model to classify land cover types from optical multiband satellite imagery, and how we packaged it as a QGIS plugin.
- Liver Segmentation from CT Scans: U-Net with Combined Dice–BCE Loss and HU Windowing Our modular PyTorch pipeline for segmenting liver regions from CT imagery — covering the U-Net architecture, Hounsfield unit windowing, combined loss formulation, and the full suite of segmentation metrics.