Isabek Altynbek

@IsabekConnected with DSML

Computer Vision Engineer | LLM/VLM Integration

Computer Vision / ML Engineer с опытом в разработке и внедрении end-to-end AI-систем в production, включая real-time компьютерное зрение, мультимодальные решения (CV + LLM/VLM) и высоконагруженные ML-пайплайны.

Open to

Hiring & referralsStartup collaboration

About

Специализируюсь на построении инженерных AI-систем полного цикла - от данных до деплоя. Реализую системы обработки видеопотоков в реальном времени с задержкой ~30 ms и стабильной работой на уровне 20 FPS, включая интеграцию моделей детекции, трекинга и LLM-логики принятия решений. Покрываю полный ML lifecycle: сбор и разметка данных (CVAT, ~30k+ объектов), обучение, оптимизация и деплой моделей в production-среду. Ключевой опыт - разработка автономной AI-системы для дрона с объединением YOLOv11, DeepSORT, OpenVINS и LLM (Qwen, GPT-4) для интерпретации команд на естественном языке и динамического планирования маршрутов. Проектирую микросервисную архитектуру (Docker, REST, gRPC), реализую real-time video pipelines, оптимизирую инференс и обеспечиваю устойчивость моделей в сложных условиях эксплуатации. Внедряю caching-слой (Redis + PostgreSQL) для ускорения принятия решений, а также системы мониторинга, логирования и experiment tracking (MLflow / аналоги) для контроля качества и воспроизводимости в production. Дополнительно разрабатываю ML-решения для анализа энергетических данных с высокой точностью прогнозирования (R2 = 0.9997, ошибка ~0.1%), а также системы сегментации промышленных объектов с точностью >90%. Работаю с табличными и визуальными данными: EDA, feature engineering, обучение и сравнение моделей (Random Forest, XGBoost и др.). Основной фокус - создание не отдельных моделей, а устойчивых AI-систем, которые стабильно работают в реальных условиях, масштабируются и дают измеримый результат в продакшене.

Career

Experience: 3 years

Experience

Computer Vision Engineer

RISE Innovations Ltd.

Develops AI systems for autonomous devices and industrial analysis, processing real-time video streams (20 FPS, ~30 ms latency) and integrating Computer Vision and LLM/VLM models for decision-making. Production-ready solutions use microservice architecture and scalable infrastructure.

Oct 2025 — Apr 2026

Built and deployed a real-time drone CV/LLM system, achieving ~20 FPS, ~30 ms latency, and up to 92% recall.

• Designed and implemented a CV + LLM/VLM drone autonomy system for real-time decision-making. Built video processing for two cameras with OpenCV, FFmpeg, GStreamer, and RTSP; translated natural-language instructions into trajectories using Qwen and GPT-4; integrated OpenVINS for navigation and dynamic route correction. Built a low-latency video pipeline and adaptive planning based on CV data.

Freelance Computer Vision / ML Engineer

Freelance

Apr 2026 — Jun 2026

Delivered solar-panel defect analysis and energy-generation forecasting; reported >90% accuracy, R2 = 0.945, and 50% less image-processing time.

• Built a YOLOv11 system to segment solar-panel defects (dust, snow, and damage), achieving >90% accuracy. • Collected and annotated a dataset of 100+ images and applied augmentation to improve model quality and robustness. • Calculated power losses based on defect area, enabling quantitative assessment of panel efficiency. • Classified panel condition as Normal, Maintenance, or Critical to support operational decisions. • Visualized results by overlaying defect annotations, power losses, and status on images. • Automated the end-to-end image-analysis pipeline, reducing processing time by 50%.

Product Data Analyst (Growth / Marketing Analytics)

Yuzhny Polevod

Agricultural technology company. The resume describes an initiative to optimize its YouTube channel, increasing organic traffic by 50% in six months without additional resources.

Sep 2023 — Jul 2025

Analyzed YouTube audience behavior and optimized posting times, contributing to 50% organic-traffic growth in six months.

• Analyzed the YouTube channel without being assigned the task, identifying opportunities to grow traffic. • Collected and processed likes, comments, and watch-time data using Python, pandas, and scikit-learn to structure behavioral metrics. • Built visualizations with matplotlib and seaborn to identify audience-activity patterns. • Identified optimal publishing times by day of the week and month, contributing to 50% organic-traffic growth over six months.

Education

Bachelor's degree, Industrial Electronics

Tomsk State University of Control Systems and Radioelectronics

2017

Machine Learning Engineer

Skillbox

2024 — 2025

Selected projects

Autonomous AI system for a drone

Real-time computer vision, object detection and tracking, visual navigation, and LLM integration.

Geological core segmentation

Developed a system to segment and analyze geological images using deep-learning methods.

Solar-panel defect detection

Automated detection of dirt and damage on solar panels, with estimated power losses.

Electricity-generation forecasting

Developed machine-learning models to forecast energy generation from meteorological data.

Skills

Computer VisionMachine LearningDeep LearningArtificial IntelligencePythonPyTorchOpenCVYOLOv11DeepSORTObject DetectionImage SegmentationInstance SegmentationVideo AnalyticsReal-Time Computer VisionLLM IntegrationVision-Language Models (VLM)Multimodal AICV + LLM SystemsRTSPFFmpeg