Skip to content
KA

Karina Abubakirova

@karinaa-abConnected with DSML. This person's Telegram account has been confirmed through DSML Kazakhstan.

AI/ML Research Engineer

AI/ML Research Engineer with an MSc in Computer Vision, specializing in multimodal learning and retrieval-augmented systems.

Open to

Project collaborationResearch collaborationStartup collaborationLearning with othersMentoring others

About

First-author researcher in vision-language models with proven improvements on benchmark datasets (YouCook2, Ego4D-HCap). Experienced in building and evaluating end-to-end ML pipelines using Python and PyTorch.

Career

Experience: 2 years

Experience

Graduate Researcher (Computer Vision)

Mohamed Bin Zayed University of Artificial Intelligence

A graduate-level AI university based in Abu Dhabi.

Aug 2024 — Present

Built retrieval-augmented and multimodal video-captioning methods, reporting up to 11.1% CIDEr improvement; first author on four papers.

• Built and evaluated a retrieval-augmented video captioning pipeline using ConceptNet-derived knowledge and reliability-aware gating, improving robustness on YouCook2 and Ego4D-HCap benchmarks. • Developed a cross-modal fusion transformer integrating visual features and retrieved knowledge, achieving up to 11.1% CIDEr improvement. • Designed a perturbation-based evaluation framework (Best–Worst, TinyLie, SpanFix) to assess factuality and reduce hallucinations in vision-language models. • First author on 4 papers in multimodal and retrieval-augmented learning (3 accepted, 1 under review).

Machine Learning Intern

Home Credit Bank Kazakhstan - Risk Department

Home Credit is an international consumer finance group; this role was in its Kazakhstan risk department.

Jul 2025 — Sep 2025

Evaluated risk-model stability and predictors, and automated monthly model updates and scoring.

• Evaluated production risk-model stability with PSI, Gini, IV, and KS, leading to early detection of predictor drift. • Identified and removed unstable predictors through marginal contribution and stability checks. • Built a monthly ETL pipeline that automatically updates and scores models on the 15th of each month. • Tested models across train, validation, test, and OOT splits to reduce risk of performance degradation.

Business and Systems Analyst

Arcana Inc.

Aug 2022 — Jul 2024

Translated stakeholder needs into technical specifications and API documentation, and supported software users and project processes.

• Facilitated communication between stakeholders and developers, ensuring project objectives were met. • Developed technical specifications and API documentation for web/app development teams. • Created end-user manuals to improve adoption and understanding of software functionalities. • Streamlined business processes to increase efficiency across projects.

Mathematics Tutor

TerenOi

Feb 2020 — Sep 2021

Tutored high school students for Unified National Testing and tailored lessons using student-progress tracking.

• Coached high school students for Unified National Testing. • Designed and delivered lesson plans tailored to student needs. • Tracked student progress in CRM and addressed learning gaps through targeted feedback.

Community Manager

Totem and Bas.app

Represented two startup apps at a digital forum, supporting onboarding and gathering user feedback.

• Represented the Totem and Bas.app startup mobile applications at Almaty’s Digital Forum, assisting with user onboarding, demonstrating key app features, and collecting user feedback for improvement.

Education

MSc, Computer Vision

Mohamed Bin Zayed University of Artificial Intelligence

2024 — 2026

BSc, Information and Communication Technologies

Kazakh-British Technical University

2019 — 2023

Erasmus+ International Credit Mobility, Exchange Student

Coimbra Higher Institute of Engineering

2022 — 2022

Selected projects

FoodGen - Vision-Language Modeling

Fine-tuned InstructBLIP on Food101 and Food500Cap; built an automated evaluation pipeline for caption generation.

Real-Time Facial Keypoint Detection

Developed and deployed a deep learning model for facial landmark detection and emotion classification in real-time applications.

Image Classification System

Built and optimized CNN models on ImageWoof, improving classification accuracy through architecture tuning.

Alzheimer’s Disease Prediction from Gut Microbiome Data

Developed machine learning models to identify microbial biomarkers associated with Alzheimer’s disease, building a non-invasive diagnostic pipeline using structured biological data.

KBTU Mobile App (iOS)

Developed a production mobile application using Swift and Firebase with authentication and real-time data integration.

Achievements

Selected for a fully-funded academic exchange program under Erasmus+

Erasmus+ International Credit Mobility · 2022

Skills

Computer VisionMultimodal LearningRetrieval-Augmented ModelsModel EvaluationFeature EngineeringPythonC++SQLPyTorchscikit-learnNumPypandasGitJupyterMatplotlib