Aidar Amangeldi

@aidarjpgConnected with DSML

ML Engineer with expreience in CV and NLP

Machine Learning Engineer with 4+ years of experience developing AI solutions across healthcare, computer vision, natural language processing, and embedded AI.

Open to

Hiring & referralsPaper discussionsProject collaborationResearch collaboration

About

Experienced in designing, implementing, and deploying production AI systems using Python, deep learning, Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), and modern MLOps practices. Strong background in AI system integration, data engineering, model optimization, and scalable machine learning infrastructure. IEEE-published researcher with expertise in translating cutting-edge AI research into production-ready applications.

Career

Experience: 4 years

Experience

AI Engineer

Medical Center of the President’s Affairs Administration of the Republic of Kazakhstan

Apr 2025 — Present

Develops clinical LLM and RAG applications and data pipelines, deploying AI services in Kubernetes environments.

• Developed enterprise LLM and Retrieval-Augmented Generation (RAG) applications for physicians, pharmacists, and dietitians, building Python backend services, REST APIs, and vector-database-backed semantic retrieval for context-aware clinical recommendations. • Built data ingestion, preprocessing, and transformation pipelines integrating structured and unstructured clinical data through HL7 v2, FHIR R4, DICOM, and SQL-based data sources.

Research Assistant – Embedded Systems & AI (ESAI) Laboratory

Nazarbayev University

Aug 2022 — Present

Researches efficient deep learning on edge hardware; developed a DVFS algorithm that reduced fine-tuning energy use by up to 37% and co-authored two IEEE publications.

• Fine-tuned and benchmarked transformer language models (BERT, DeBERTa, Pythia, SmolLM) and CNN/Vision Transformer models for deployment on resource-constrained edge hardware (NVIDIA Jetson AGX Orin), optimizing energy, latency, and accuracy trade-offs. • Designed and implemented a GPU utilization-guided Dynamic Voltage and Frequency Scaling (DVFS) algorithm that reduced fine-tuning energy consumption by up to 37% with near-zero profiling overhead, generalizing across 7+ model architectures and multiple GLUE benchmark tasks (SST-2, QNLI, MRPC, RTE, CoLA).

Data Science Intern

Institute of Smart Systems and Artificial Intelligence (ISSAI)

May 2022 — Aug 2022

Helped build a Central Asian dishes image dataset and trained food-classification models, reaching 80% accuracy.

• Contributed to the development of the first Central Asian dishes dataset (published in MDPI Nutrients), web-scraping over 6,000 images across 10 dish categories and performing iterative annotation and quality control. • Trained YOLO and ResNet models on 8 food classes, achieving 80% classification accuracy and 0.83/0.80/0.87 precision/recall/F1.

Education

Master of Science, Data Science

Nazarbayev University

2024 — 2026

Bachelor of Science, Computer Science

Nazarbayev University

2020 — 2024

Selected projects

Workforce Layoff Forecasting & Analytics

Built classification, regression, clustering, and time-series workflows to forecast workforce reductions and labor-market trends for planning and decision support.

AI-Generated Text Detection with Topic-Based Generalization

Developed NLP pipelines and evaluated classical and transformer-based models for AI-generated text detection and robustness across unseen topics; achieved 88.9% classification accuracy and 0.96 ROC-AUC.

Efficient Vision Transformer Optimization

Benchmarked CNN and Vision Transformer architectures on Tiny ImageNet and DermaMNIST, with up to 75% fewer model parameters while maintaining competitive classification performance.

Multimodal Congenital Heart Disease Diagnosis

Developed deep-learning pipelines combining chest X-ray images and heart-sound recordings; ensemble models achieved 80.7% image classification accuracy.

Skills

PythonMLCVNLPMath