Skip to content
Yerdauit Torekhan

Current research

Under reviewImage tokens inside vision–language decoders, compared with human visual cortex
ManuscriptDetection limits of topological analysis for grid-cell codes in world models
ActiveTime perception in spaceflight environmentsIf the entorhinal system encodes space and time in a shared format, distorting one should be legible in the other.Details In designA controlled comparison of colour harmony modelsThe classical rules of colour harmony are a century-old doctrine that has barely been tested on people.Details

Publications

Teaching

Fall 2026

Introduction to Machine Learning (CSCI3237)

Tutor — five lab groups, 120 students

Running the practical sessions for the undergraduate machine learning course — three credits over fifteen weeks.

Kazakh-British Technical University

Spring 2026

Web Development (INFT2205)

Practice Instructor

Taught the practical sessions for the undergraduate web development course — building, reviewing and grading student projects.

Kazakh-British Technical University

Spring 2026

Object-Oriented Programming (CSCI2106)

Practice Instructor

Taught the practical sessions for the undergraduate OOP course — weekly problem sessions on object-oriented design and debugging, lab reviews and code assessment.

Kazakh-British Technical University

Experience

  1. Lead Computer Vision Researcher

    Apr 2026 – Present

    Qoima Automation · Almaty

    • Lead the design of an industrial computer-vision platform from scratch — real-time multi-object detection, tracking, and structured video understanding — directing a small engineering team.
    • Architected a prediction-error-driven event-segmentation pipeline in which salient moments trigger VLM analysis of video clips, feeding a queryable knowledge graph.
    • Designed the edge–core deployment split — on-device real-time detection, centralized video reasoning — including GPU selection and the end-to-end data pipeline.

  2. Computer Vision Researcher — Face Anti-Spoofing

    Aug 2025 – Apr 2026

    OZ Forensics · Almaty

    • Shipped an end-to-end face anti-spoofing pipeline (collection, training, evaluation, deployment) into a production system processing 100M+ identity verifications per month.
    • Replaced ad-hoc accuracy reporting with threshold-aware evaluation tooling (FAR, FRR, FAR@FRR analysis), enabling meaningful model comparison.
    • Benchmarked and fine-tuned vision backbones across attack types and demographic slices, identifying where learned representations degrade under distribution shift.

  3. Research Engineer

    Feb 2024 – Present

    FLIS Lab, KBTU — Fuzzy Logic and Intelligent Systems (PI: Prof. P. Shamoi) · Almaty

    • Designed the visual-feature extraction pipeline and evaluation protocol for a first-author study on painting aesthetics, published at IEEE SIST 2025.
    • Co-authored 4 further papers on multimodal emotion modeling and fuzzy color representation, contributing experiment design, model implementation, and manuscript writing.
    • Drafted methodology and results sections of funded grant proposals submitted by the lab.

  4. Junior Computer Vision Engineer

    Sep 2024 – Aug 2025

    BTS Digital · Almaty

    • Developed face anti-spoofing models for real-world attack scenarios and evaluated them with APCER, BPCER, and ACER per ISO/IEC 30107-3.
    • Built a retrieval-augmented generation system (LangChain, custom tokenization, vector search) coupled to a DeepSeek generator for domain-specific question answering.
    • Automated data-generation workflows in ComfyUI, expanding training corpora for downstream CV models.

  5. Data Engineer

    Apr 2024 – Sep 2024

    Adata · Almaty

    • Built ETL pipelines in Talend and Python, refactored hot paths for throughput, and added pandas validation routines that cut downstream errors.

Education