Koyilbek Valiev
AI / ML Engineer | Research Engineer
AI / ML Engineer with 3+ years of experience in AI agents, computer vision, vision-language models, and end-to-end ML pipelines. Currently building autonomous AI agents at TelecomSoft in Tashkent. Experienced in video anomaly detection, time series forecasting, multimodal retrieval, agentic AI systems, ASR/TTS integration, experiment tracking, and scalable intelligent system design.
News & Updates
- Jan 2026 Joined TelecomSoft in Tashkent as AI / ML Engineer, building autonomous customer support agents
- Aug 2025 2nd Place at HumbleBeeAI Hackathon: Data Science Challenge
- Mar 2025 Started co-op at Pia Space, Seoul working on multimodal AI for video anomaly detection
- Feb 2024 Promoted to AI Team Lead at Recs Innovation, leading solar forecasting and anomaly detection projects
- Dec 2023 Won 1st place in Woosong University Capstone Competition among 200 teams
- 2024 Filed two patent applications for smart grid technology; one issued by KIPO
Experience
Jan 2026 –
Present
Present
AI / ML Engineer Current
TelecomSoft, Tashkent, Uzbekistan
- Built autonomous AI agents from scratch that automate customer support with high accuracy and human-style conversations, leveraging external tool use and multi-skill orchestration
- Integrated ASR and TTS pipelines end-to-end, enabling natural voice-based agent interactions with support for diverse accents and dialects
- Developed multi-agent architectures with function calling, tool use, and orchestrated workflows for complex customer queries
- Designed NLU modules capable of understanding varied chat styles, accents, and multilingual customer inputs
- Deployed open-source models to production, built API-connected backend services, and developed web-based UI for real-time agent monitoring and interaction
- Designed benchmarks and evaluation pipelines to assess AI agent workflows, measuring accuracy, latency, and end-to-end performance across different scenarios
- Built observability and tracing infrastructure to track agent decisions, tool calls, and conversation quality throughout the entire pipeline
Mar 2025 –
Jun 2025
Jun 2025
Multimodal AI Engineer (Co-op)
Pia Space, Seoul, South Korea
- Collected and annotated diverse anomaly video datasets (violence, falls) using CVAT and proprietary tools
- Optimized lightweight VLMs (<1B params) for video anomaly classification with dynamic sampling and prompt engineering in few-/zero-shot settings
- Co-developed MACS 3.0, a prompt-driven multimodal system for CCTV anomaly detection, improving accuracy by ~24%
- Built benchmark dataset with manually captioned videos for cross-modal retrieval (video-to-text, text-to-video, video-to-video)
- Developed and evaluated cross-modal retrieval models to enhance scalable multimodal video understanding
- Implemented multimodal explanation maps for VLMs, generating visualizations that highlight key words and regions
Feb 2024 –
Mar 2025
Mar 2025
AI Team Lead | AI Engineer (Co-op)
Recs Innovation Ltd, Naju, South Korea
- Built ML pipeline (cleaning, anomaly detection, feature engineering, clustering) reducing bidding prediction error to 0.0017% and increasing bid win rate by 15x
- Developed full-stack 48-hour solar power forecast using LSTM-CNN, GRU, Transformer models; achieved 6% error, deployed in Sun-Q EMS
- Led unsupervised PV anomaly detection (LSTM Autoencoder, LSTM-VAE, TranAD, VAE) achieving >70% F1-score on real-world data
- Authored technical documentation; managed AI team using Notion and ClickUp
- Wrote two patent applications for smart grid technology
Sep 2023 –
Dec 2023
Dec 2023
AI Engineer (Capstone — NetVision)
Woosong University, Daejeon, South Korea
- Coordinated team of 4 engineers to develop a real-time trash-bag optimization system for NetVision
- Developed YOLOv8 to detect, classify, segment, track & size-estimate bags in video streams
- Won 1st place among 200 teams in Woosong University Capstone Competition
Jun 2023 –
Aug 2023
Aug 2023
Computer Vision Intern
Sequus PTY LTD, Australia (Remote)
- Automated conversion of bounding-box outputs into YOLO, Pascal VOC, and custom formats
- Annotated hundreds of architectural drawings (LabelImg) to improve model accuracy
Research & Publications
A Study of Reinforcement Learning Framework for Energy Management in Smart Grids: Integrating Market Trading, Load Forecasting, and Vertical Agents
KICS, Vol. 51, No. 2, pp. 324–347, Feb. 2026 — DOI: 10.7840/kics.2026.51.2.324
Patents
ESS를 활용한 에너지 비용 및 판매 가격 계산 방법 및 장치
Inventor: Koyiljon Valiev — Recs Innovation Ltd, South Korea — Korean Intellectual Property Office (KIPO)
Smart Grid Technology — Patent Application II
Recs Innovation Ltd — Intelligent energy management system
Skills
AI & ML Frameworks
Agentic AI & LLM Tools
Computer Vision
Data Science & Analysis
DevOps & Infrastructure
APIs & Prototyping
Education
Sep 2021 –
Aug 2025
Aug 2025
B.Eng. in Artificial Intelligence & Big Data
Woosong University, Daejeon, South Korea
Coursework: Artificial Intelligence, Machine Learning (Theory & Lab), Computer Vision, Data Visualization, Statistics, Linear Algebra, Discrete Mathematics, Capstone Project.
Achievements & Awards
- 2nd Place — HumbleBeeAI Hackathon: Data Science Challenge (Aug 2025)
- 1st Place — Woosong University Capstone Competition among 200 teams (Dec 2023)
- President's Award — Woosong University (Dec 2023)
- 100% Merit Scholarship — Woosong University (Sep 2023 – Dec 2023)
- 1st Place — IoT Learning Concert (Sep 2023 – Dec 2023)
- 2nd Place — Machine Learning Lab Learning Concert (Sep 2023 – Dec 2023)
- Korean Patent Issued — ESS-based energy cost & sales price calculation method, KIPO (2024)