Skip to content
ST

Sungkar Turmagambetov

@SungkarConnected with DSML. This person's Telegram account has been confirmed through DSML Kazakhstan.

Computer Vision Engineer

MSc student in Visual Computing at TU Wien (Vienna) with production experience as an AI Engineer and Data Scientist.

About

I build computer vision systems end-to-end — from detection pipelines and 3D reconstruction to deployed ML services. Background spans deep learning, classical CV, and MLOps. Originally from Kazakhstan, currently based in Vienna.

Career

Experience: 1 years

Experience

Data Scientist; Artificial Intelligence Engineer

Kazakhtelecom JSC

Kazakhtelecom is a telecommunications provider in Kazakhstan.

Aug 2024 — Sep 2025

Built churn models and retention analysis, then deployed chatbots and document automation that reduced operator workload and approval errors.

• Predictive Modeling: Churn model (82% ROC-AUC, Scikit-learn) on B2B/B2C/B2G data – retained ∼5% more subscribers in pilot. • Customer Analytics: Feature engineering (Pandas, SQL); churn driver analysis informing retention campaigns across all segments. • Process Automation: Built and deployed production-scale AI chatbots, cutting operator workload by ∼30%. • Document Intelligence: Automated verification/proofreading agents (Hugging Face NLP) – reduced errors by 60%, approval time 3 days → 1. • MLOps: Model monitoring (prediction drift, failure rates) for deployed AI services in production.

Education

MSc, Visual Computing

Technische Universität Wien

2026 — 2028

Bachelor, Automation and Control Engineering

Almaty University of Power Engineering and Telecommunications

2020 — 2024

Selected projects

Cross-View 4D Novel View Synthesis

Given a monocular video stream, synthesised renders from arbitrary viewpoints at arbitrary timesteps, combining dynamic 3D reconstruction with novel view synthesis over a 4D space-time representation.

Visualising Trade Disruption: Flight Data vs. Economic Indices

Interactive geospatial dashboard correlating real AIS/flight movement data with global economic indices, revealing how conflict zones reshape trade corridors and flight patterns. Animated TripsLayer, FDEB edge bundling, time-series overlays; FastAPI and SQLite backend on Oracle Cloud.

License Plate Recognition (LPR) System

Real-time two-stage pipeline: YOLOv5 for localisation and CRNN with CTC Loss for variable-length OCR without segmentation. Optimised for live video stream inference.

Automatic Image Colorization

Custom Weighted L1 Loss improved saturation in complex regions by ∼25%; LAB color space preprocessing with L-channel input and ab-channel targets.

Skills

Object Detection (YOLO)SegmentationNovel View Synthesis3D ReconstructionCamera GeometryPyTorchTensorFlowResNetU-NetViTCRNNNeRFtrainingfine-tuningablation studiesOpenCVPandasSciPyMatplotlibDocker