Baisal Tursynkhan

@baisal

Junior Machine Learning Engineer, Demand Forecasting

Fourth-year CS student at Nazarbayev University and an ML Engineer, where I build demand forecasting models for an online grocery service.

About

Most of my work is time series: a global LightGBM model across 1,000 SKUs, separate methods for intermittent demand, and promo features. Interested in forecasting, applied ML in supply chain, and getting models to actually work in production.

Career

Experience: 1 years

Experience

Machine Learning Engineer

Arbuz.kz

Online grocery retailer and delivery service in Kazakhstan.

Mar 2026 — Present

Built demand forecasting models and an out-of-stock cause classifier, improving WMAPE and cutting daily analysis time for buying teams.

• Reduced forecast WMAPE by 15% against the baseline across sales for 1,000 SKUs and 8 warehouses by building a global LightGBM demand model with a 30-day horizon, validated using rolling-origin backtesting. • Reduced WMAPE by 13% for 300 intermittent-demand SKUs by separating them into a dedicated track using Croston SBA/TSB and AutoETS through StatsForecast where boosting performed worse. • Reduced WMAPE during promotional periods by 30% using promotion features built from promotion history reconstructed in dbt.

Research Assistant, Behavioral Finance

Nazarbayev University

Research university in Kazakhstan.

Nov 2025 — Jul 2026

Built data collection and event logging systems for behavioral finance experiments involving more than 2,000 students.

• Collected decisions from more than 2,000 students across 8 experimental branches by designing backend logic and a data collection system in oTree, with validation through automated bot tests. • Implemented real-time recording of every user choice change in a binary decision tree using a JavaScript event logging system with input validation and state transitions, without third-party frameworks.

Backend Developer

Kaspi Analytics

Jun 2024 — Aug 2024

Developed backend services and WhatsApp bot features that improved response times, automated notifications, and increased seller page reviews.

• Reduced customer response time by 25% by developing Flask backend services to process API requests between a WhatsApp bot and customers. • Automated 19% of customer notifications about order status and delivery while contributing to a WhatsApp bot built with Twilio. • Increased the number of reviews on seller pages by 20% by creating a service that sends customers parcel information along with a request to leave a review.

Education

Bachelor's degree, Computer Science

Nazarbayev University

2022 — 2027

Selected projects

Dynamic Pricing System

Developed an ML service to forecast demand and optimize prices for expected profit, accounting for seasonality and competition. Trained LightGBM and XGBoost models to estimate price elasticity and used SHAP values to make model decisions understandable to business users. Designed a high-load REST API with FastAPI, containerized it with Docker, and integrated Redis for fast retrieval of historic...

Cybersecurity Platform

Developed a cybersecurity monitoring platform with a microservices architecture, real-time log analysis, and ML-based anomaly detection to identify threats and suspicious activity. Built backend services with FastAPI and Docker and a React/TypeScript frontend with interactive security dashboards, log processing, and threat analytics. Integrated ML anomaly detection and AI threat analysis for pr...

MyCar

Built an AI tool for automated resume screening and generating initial candidate questions using the OpenAI API, simplifying recruiters' work.

Achievements

4th place among more than 50 teams at HackNU

HackNU · 2025

Skills

PythonGoJavaC++TypeScriptJavaScriptSQLFastAPIFlaskDjangoWailsReactSpring BootGitPostmanDockerRedisdbtClaude Code Skillspandas