Hackathon project. Turns fund bank-statement rows into journal entries and shows the evidence behind every proposal. Ten matcher stages propose a counterparty, project code, deal, position and classification for each row, automated checks catch problems such as broken running balances, and a review queue hands a person only what isn't clear-cut. On the sample data it found 23 of 100 rows where the administrator's own documented rule contradicts the file they shipped.

Predicts apartment prices from a public listings dataset (7,565 listings, 7,391 after cleaning) following the CRISP-DM process. One-hot encoded 12 districts and compared a Linear Regression baseline (R² 0.73) with a Random Forest (R² 0.79, MAE $9.8k). District alone drives up to a 3.2x price difference between the cheapest and most expensive areas.

Multiclass prediction of patient status (Status_C, Status_CL, Status_D) for a Mohirdev / Kaggle competition, optimised for multi-class log loss. Reproducible notebook: feature engineering on tabular medical data, stratified 5-fold cross-validation, LightGBM, XGBoost and CatBoost models, and a blend of their probabilities exported in the required submission format.

Diploma project. An ensemble of 7 models (LSTM, XGBoost, ARIMA, Prophet, linear regression, decision tree, SVM), weighted by performance, forecasting S&P 500 prices (RMSE $12.62, MAE $10.90). Built a walk-forward backtesting engine that tracks Sharpe ratio, max drawdown, win rate and total return across 500+ tickers. Directional accuracy was only about 29% on 34 forecasts, in line with the efficient-market hypothesis, so the real result is the evaluation framework, not a trading edge. Presented at the university poster session.

Built a comprehensive multilingual wedding planning platform with React frontend and Node.js backend. Features role-based access control, guest management, RSVP system, photo galleries, guest book, and real-time WebSocket updates. Supports English, Uzbek, and Russian languages with PostgreSQL database and JWT authentication.

Built a complete meal prep and delivery management system for a health-focused food business. Features personalized nutrition plans, goal tracking, subscription management, and multi-language support (Uzbek, Russian, English). Includes customer intake system, nutrition calculator, meal selection from weekly menu, admin panel, kitchen dashboard, and Click payment integration.

Developed a full-stack grocery delivery web application with React frontend and Node.js backend. Features include multi-language support (5 languages with RTL), real-time shopping cart, role-based dashboards (customer, admin, super admin, rider), interactive map-based address selection, order management, analytics, and secure authentication. Built with PostgreSQL, Drizzle ORM, WebSocket for real-time updates, and a mobile-first responsive design.

Designed and developed a professional personal website for Firdavs Esaev, a lawyer in Uzbekistan, using React, TypeScript, and Tailwind CSS. Created a modern, responsive design with multi-language support (Uzbek, Russian, English), integrated booking system, admin dashboard, and contact management. Features include client testimonials, professional credentials, and flexible appointment scheduling.

Built a complete educational platform with React frontend, Node.js backend, and multi-role dashboard system. Implemented internationalization, analytics tracking, and responsive design. Successfully deployed to production with GitHub integration.

Built a comprehensive agricultural investment platform connecting farmers with investors. Features multi-language support (English, Russian, Uzbek), project listings, funding tracking, and community features. Implemented secure authentication, project management, and real-time collaboration tools for sustainable agricultural growth.