This project focuses on building an end-to-end time series forecasting system to predict daily grocery sales for different product families across multiple stores. The goal is to leverage historical sales data along with external factors such as promotions, holidays, and oil prices to accurately forecast future demand.

Automated domain-investing pipeline · AI Council valuation (Claude + DeepSeek) · hard budget walls · 22-case end-to-end smoke · MIT

Experimented with 16+ deep neural network configurations in TensorFlow on e-commerce demand data to compare tuning sensitivity, performance, and runtime tradeoffs. Best model: 2-layer (128->64), ReLU, Adam, LR 0.001, batch 128, early stopping, validation MAE 12.54 in ~126 seconds.

Расчет шпунтового ограждения по ВСН 136-78 с автоматическим подбором глубины заглубления, эпюрами активного/пассивного давления, изгибающего момента и формированием инженерного отчета.

Metadata-first creative asset pipeline demonstrating workflow automation, schema governance, and production efficiency optimization.

ExtendScript-based Photoshop automation tool that programmatically creates structured artboards with metadata-driven naming and layout logic.

A pattern (not a framework) for AI-agent pipelines — agents + validators + manifest, validated across 6+ production projects