GitHub | Google Scholar | StackOverflow | LeetCode | LinkedIn

AI/ML & MLOps Engineer

AI/ML & MLOps Engineer with 6+ years of industry and research experience focused on Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), Vision-Language Models (VLMs), and scalable MLOps infrastructure. Skilled in fine-tuning and deploying transformer architectures, building Kubernetes-based ML platforms, and automating model training and deployment across cloud and on-prem environments.

Currently working on designing and maintaining private cloud and Kubernetes clusters for distributed training and inference workloads. Building systems for resource discovery, policy enforcement, and workload scheduling using Kubeflow, Kueue, Kai Scheduling, and Kyverno, while focusing on scalable, reliable, and automated MLOps pipelines.

Experienced in developing and optimizing LLM-based, multimodal, and retrieval-augmented systems with strong background in model optimization (TensorRT, Triton Inference Server, vLLM), CI/CD integration, and production-grade deployment across AWS and GCP.

Skills Summary

  • Programming Languages: Python, C/C++, Java
  • Data & Query Systems: MySQL, PostgreSQL, PySpark; data wrangling with Pandas, Polars; visualization using Power BI
  • ML & DL Frameworks: NumPy, Scikit-learn, PyTorch, PyTorch Lightning, TensorFlow, Keras, Hugging Face, LangChain, Unsloth
  • Retrieval & Indexing Systems: FAISS, Pinecone, Elastic, Tantivy (lexical, semantic, hybrid search)
  • MLOps & Deployment: Docker, FastAPI, Flask, gRPC, TorchServe, Triton Inference Server, TensorRT, vLLM, Kubernetes, Kubeflow, Helm, ArgoCD, Kueue, Kai Scheduling, Kyverno
  • Cloud Platforms: AWS (EC2, S3, Lambda, SageMaker, EKS), Google Cloud Platform (Vertex AI, GKE, Compute Engine, Cloud Run)
  • Core Competencies:
    • LLMs & RAG: LoRA/PEFT fine-tuning, retrieval pipeline design (Sentence Transformers, Tantivy), vLLM-based inference orchestration, knowledge base integration
    • MLOps & Infrastructure: Cluster orchestration, workload scheduling, resource management, model lifecycle automation, CI/CD, observability (Prometheus, Grafana, ELK)
    • Conversational AI: Voice-to-voice AI agents with ASR, LLMs, and TTS pipelines using LiveKit and related tools
    • Multimodal AI: Vision-Language Models, OCR, object detection, tracking, re-identification, video action recognition
    • Generative AI: sLLMs, Diffusion Models, GANs, DeepFakes, image-to-video, face restoration and enhancement
    • Edge Deployment: Model pruning, quantization, on-device inference, latency optimization
    • ML Engineering: End-to-end pipeline design, scalable deployment, model monitoring, MLOps best practices

Work Experience

MLOps Engineer

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

  • Kubernetes & Private Cloud Platform: Designing, deploying, and maintaining Kubernetes clusters for a full-featured private cloud platform supporting distributed model training, inference, and MLOps workloads. Building multi-tenant infrastructure with automated resource provisioning, monitoring, and scaling for GPU-intensive pipelines.
  • Resource Discovery & Policy Optimization: Implementing dynamic resource discovery and scheduling strategies to maximize GPU and CPU utilization. Developing custom resource management operators for efficient allocation and preemption, and integrating policy optimization for fair and performance-aware workload balancing.
  • Policy Enforcement & Admission Control: Establishing cluster-wide governance using Kyverno and custom admission webhooks. Automating security and compliance enforcement, applying mutation/validation logic, and standardizing deployment configurations across namespaces and teams.
  • Advanced Scheduling & Job Queueing: Integrating and enhancing Kueue and Kai Scheduling frameworks to improve distributed job scheduling and workload queueing efficiency, achieving better throughput and fairness for model training and data processing jobs.
  • Kubeflow & MLOps Pipelines: Managing the full lifecycle of ML pipelines using Kubeflow, including Katib for automated hyperparameter tuning, pipeline orchestration, and deployment. Leveraging Helm and ArgoCD for continuous delivery of ML workloads across staging and production environments.
  • Cluster Automation & Observability: Automating cluster provisioning, logging, and monitoring pipelines using Prometheus, Grafana, and the ELK stack. Establishing CI/CD integration for infrastructure and ML workflow updates using GitOps principles.

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 the Korean language using LoRA and DPO (Direct Preference Optimization) for efficient on-device deployment. Also fine-tuned GPT-3.5-turbo on conversational data using custom augmentation workflows.
  • AI Agent & Conversational Systems: Built an end-to-end voice-to-voice AI assistant using LiveKit for real-time media streaming, Whisper for ASR, OpenAI-based LLMs for dialogue, and TTS for responses. Integrated multimodal understanding (face ID, age/gender, emotion). Enabled dynamic function/tool calling via LLMs to invoke external APIs using MCP servers, allowing real-time task execution and tool orchestration within conversation flow.
  • Retrieval & Data Tooling for LLMs: Built a simulated interaction tool to collect structured dialogue data for fine-tuning LLMs. Managed persistence with PostgreSQL and hosted the system on GCP.
  • LLM Inference & API Optimization: Designed scalable APIs using vLLM and FastAPI for real-time LLM inference. Applied LangChain and concurrency patterns to optimize API orchestration for latency-sensitive workflows.
  • Generative Model Training & Deployment: Fine-tuned FLUX (Stable Diffusion) using LoRA for stylized character generation. Deployed as a REST API on GCP.
  • VLM Deployment for Context Awareness: Developed a visual perception module for virtual characters using Phi-3-Vision, enabling multimodal awareness and interaction.
  • Face Parsing & Enhancement: Enhanced face-swapping pipeline by integrating a custom face parsing model and improving backbone efficiency for segmentation accuracy.
  • DeepFake & Face Restoration: Worked on face restoration and enhancement techniques to improve DeepFake video quality used in high-visibility media projects. Can be seen here.

ML Engineer

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

  • Video-Based Visual Content Moderation: Developed a robust pipeline for detecting inappropriate content in video streams using temporal action recognition models. Achieved a 10%+ improvement in accuracy by optimizing model architecture and training strategy.
  • Detection-Based Moderation Pipeline: Implemented real-time object detection and segmentation models to flag unsafe visual elements for brand safety. Integrated active learning loops, improving precision and recall by ~15%. Built a scalable end-to-end training and deployment pipeline on Kubeflow.
  • Classification-Based Moderation Framework: Designed a multi-label, multi-head classification system combining self-supervised and supervised learning. Boosted precision by ~20% on hard samples and established this architecture as the standard for visual moderation across projects.
  • Dataset Clustering with CLIP Embeddings: Leveraged CLIP vision-language embeddings to perform unsupervised dataset analysis. Applied PCA for dimensionality reduction and used KMeans and DBSCAN for clustering to identify content groups and outliers. Enabled efficient dataset curation and weak supervision strategies.
  • Model-Assisted Labeling System: Built a feedback-driven pipeline that combined inference on labeled and unlabeled data to accelerate data annotation. Used active learning to improve labeling quality and reduce manual annotation time.

AI Research Engineer

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

  • Slab Text Recognition: Developed an OCR pipeline to recognize handwritten text on slab metals using Spatial Transformer Networks (STN) and sequential models. Built the full workflow from data preprocessing to training and evaluation. Achieved over 90% accuracy by integrating state-of-the-art text detection and recognition techniques optimized for industrial scene images.
  • Automatic Number Plate Recognition (ANPR): Designed an end-to-end pipeline for number plate detection and recognition. Improved performance by ~15% in precision and recall through active learning, synthetic data generation, and targeted augmentation strategies.
  • Real-Time ANPR Inference: Deployed ANPR models for real-time video inference from RTSP streams, enabling continuous monitoring and detection in live camera feeds. Handled frame capture, batching, and stream resilience for production environments.
  • Car Damage Detection: Built and deployed a lightweight car damage detection model optimized for Android devices using TorchScript. Achieved a 10% improvement in precision through hyperparameter tuning and efficient model design.
  • Shadow Removal using Pix2Pix GAN: Applied Pix2Pix GAN to remove shadows cast on vehicles in captured images, enabling clearer downstream detection and damage assessment. Improved image quality and model robustness in low-light or occluded conditions.

Research Experience

Research Assistant

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

  • Computer Vision based Fire and Smoke Detection: Designed and implemented a dilated CNN architecture for improved feature extraction and recognition in images/videos. Applied optimization techniques to reduce false positives and increase inference speed by 1.5× over baseline.
  • Model Optimization for Edge Devices: Improved the FPS on Edge device (Raspberry PI 2) by using hyper-parameter tuning and quantization for detection model.

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

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


Last Updated: 2025-11-10

📥 Download Resume (PDF)