Yakhyokhuja Valikhujaev

AI/ML & MLOps Engineer

Download PDF

Summary

AI/ML & MLOps Engineer with 6+ years of experience building production ML systems across LLMs, multimodal AI, and Kubernetes-based infrastructure. Experienced in model fine-tuning, retrieval systems, inference optimization, and platform engineering for distributed training and deployment across private cloud, AWS, and GCP environments.

Skills Summary

Category Tools & techniques
Programming Python, C/C++, Java
ML Frameworks PyTorch, PyTorch Lightning, TensorFlow, Keras, Scikit-learn, Hugging Face
LLM Systems LoRA/PEFT fine-tuning, RAG pipelines, LangChain, vLLM, sentence-transformer retrieval, VLM apps
Infrastructure Kubernetes, Kubeflow, Docker, Helm, ArgoCD, Kueue, Kai Scheduling, Kyverno, Prometheus, Grafana, ELK
Inference & Serving FastAPI, Flask, gRPC, Triton Inference Server, TensorRT, TorchServe
Cloud & Data AWS, GCP, PostgreSQL, MySQL, PySpark, Pandas, Polars, FAISS, Pinecone, Elastic, Tantivy

Work Experience

MLOps Engineer

Thaki Cloud Co. Ltd | August 2025 - Present | Seoul, South Korea

  • Kubernetes & Private Cloud Platform: Designed and operated Kubernetes clusters for a private cloud platform supporting distributed training, inference, and MLOps workloads across multi-tenant GPU environments.
  • Scheduling & Policy Controls: Built resource discovery, scheduling, and policy controls for GPU-intensive workloads using Kueue, Kai Scheduling, Kyverno, and custom admission logic.
  • ML Platform Delivery: Managed Kubeflow-based pipelines and GitOps-driven deployment workflows with Helm and ArgoCD for staging and production environments.
  • Observability & Automation: Improved platform reliability through cluster automation, monitoring, logging, and standardized delivery practices using Prometheus, Grafana, and the ELK stack.

AI/ML Research Engineer

Aria Studios Co. Ltd | March 2024 - August 2025 | Seoul, South Korea

  • LLM Fine-tuning & Adaptation: Fine-tuned Qwen-2.5-7B/3B-Instruct models for Korean with LoRA and DPO, and adapted GPT-3.5-turbo on conversational datasets using custom augmentation workflows.
  • Conversational AI Systems: Built a real-time voice-to-voice assistant with LiveKit, Whisper, LLM-based dialogue, TTS, multimodal perception, and function calling through MCP servers.
  • Inference & Retrieval Tooling: Designed scalable LLM APIs with vLLM and FastAPI, and built supporting data collection and retrieval workflows backed by PostgreSQL and GCP.
  • Multimodal & Generative Applications: Delivered projects spanning Phi-3-Vision, FLUX fine-tuning, custom face parsing, and face restoration for production media workflows.

ML Engineer

Pyler Co. Ltd | July 2022 - September 2023 | Seoul, South Korea

  • Video Moderation Pipeline: Developed temporal action recognition systems for unsafe-content detection in video streams, improving accuracy by 10%+ through model and training optimization.
  • Detection & Segmentation Models: Built real-time detection pipelines for brand-safety moderation and improved precision and recall by ~15% through active-learning-driven iteration.
  • Classification Framework: Designed a multi-label, multi-head classification architecture that improved precision by ~20% on difficult samples and became the standard approach across moderation projects.
  • Data-Centric ML Tooling: Used CLIP embeddings, clustering, and model-assisted labeling workflows to speed up dataset curation and reduce manual annotation effort.

AI Research Engineer

D-Meta Co. Ltd | November 2020 - July 2022 | Seoul, South Korea

  • Industrial OCR: Built an OCR pipeline for handwritten slab-text recognition using STN and sequential models, achieving 90%+ accuracy on industrial scene images.
  • ANPR Systems: Designed and deployed number-plate recognition pipelines, improving precision and recall by ~15% through active learning, synthetic data generation, and targeted augmentation.
  • Real-Time Inference: Delivered production inference pipelines for RTSP video streams with resilient frame capture and batching strategies.
  • Mobile Vision Models: Built and deployed a lightweight Android car-damage detection model with a 10% precision improvement, and applied Pix2Pix GAN for vehicle shadow removal to improve downstream quality.

Research Experience

Research Assistant

AI and SC Lab | Sep 2018 - Nov 2020 | Seongnam, South Korea

  • Fire & Smoke Detection: Designed a dilated-CNN architecture for video-based fire/smoke detection, cutting false positives and improving inference speed 1.5× over baseline.
  • Edge Optimization: Tuned and quantized the detection model for Raspberry Pi 2, improving on-device FPS for real-time use.

Education

Institution Degree Duration
Gachon University MSc in Computer Engineering; advised by Prof. Young Im Cho; CGPA: 4.0/4.5 Sep 2018 - Feb 2021
Tashkent University of Information Technologies BSc in Computer Engineering; CGPA(%): 85/100 or 3.72/4.0 Sep 2014 - Jun 2018

Publications

  • Valikhujaev Y†, Muksimova S†, Umirzakova S, Baltayev J, Cho YI. GazeCapsNet: A Lightweight Gaze Estimation Framework. Sensors, 2025; 25(4):1224. https://doi.org/10.3390/s25041224. † These authors contributed equally to this work.
  • Valikhujaev Y, Abdusalomov A, Cho YI. Automatic Fire and Smoke Detection Method for Surveillance Systems Based on Dilated CNNs. Atmosphere, IF 2.9. 2020; 11(11):1241. https://doi.org/10.3390/atmos11111241.
  • Muksimova Sh†, Valikhujaev Y†, Cho YI. Automatic Fire and Smoke Detection System for Open Street CCTV Systems in Smart City Platforms. Korean Society of Information Scientists and Engineers, 412-414 pages, Domestic Conference.

Honors

  • Best paper award from Fire Investigation Society of Korea (FISK); (Domestic Conference, 2020)
  • Best presentation award from ISIS2019 & ICBAKE2019; (Domestic Conference, 2019)

Languages

Language Proficiency
English Full Professional Proficiency (C1 Advanced)
Korean Limited Working Proficiency (B1 Pre-Intermediate)
Russian Limited Working Proficiency
Uzbek Native Proficiency