Damir Kanymkulov

AI Team Lead

ForteBank

A major commercial bank in Kazakhstan, serving retail and corporate customers across the country's financial market.

Built and deployed seven AI agents and an SME credit-risk model, reporting 91% F1, about 20 FTE saved, and leadership of a 10-person team.

• Architected and built 7 production AI agents (OCR + LLM) automating accounting document verification and loan-utilization compliance, achieving 91% F1 score and saving ~20 FTE • Deployed agents as Docker/Kubernetes microservices integrated with Axapta ERP and Camunda BPM • Built GitLab CI/CD pipelines (lint, dev/prod releases) and Prometheus/Grafana performance monitoring • Developed an SME credit-risk (PD) model (CatBoost, LogReg, reject inference, Optuna) with an MLflow-tracked pipeline, achieving 0.57 Gini (0.79 AUC) powering profitability-based risk grading of the loan portfolio

Data Analyst

Business & Technology Services (ERG)

ERG is a global metals and mining company with operations in Kazakhstan and other countries, producing a range of metals and minerals.

Deployed an Azure RAG chatbot over ClickHouse for 20,000+ employees and automated ETL pipelines with Airflow and Greenplum.

• Deployed a RAG-based AI chatbot on Azure over ClickHouse serving 20K+ employees for real-time Q&A • Automated ETL pipelines with Airflow DAGs on Greenplum, visualized via Apache Superset dashboards

Data Scientist

Heider Lab

Improved multimodal model F1 by 20% and sentiment-analysis accuracy by 15% using optimization and BERT fine-tuning.

• Enhanced F1 by 20% in a multimodal learning framework via CMA-ES and Optuna on Azure • Improved sentiment-analysis accuracy by 15% by fine-tuning BERT on Azure GPU compute

Computer Vision Engineer

Cleverest Technologies

Deployed YOLOv8 object detection and DeepFace recognition as Docker services on Kubernetes, with detection mAP of 0.65.

• Deployed YOLOv8 detection (mAP = 0.65) and DeepFace recognition as Docker services on Kubernetes

Data Scientist

Nazarbayev University

A research university in Kazakhstan, with undergraduate and graduate programs across engineering, science, and other disciplines.

Improved monocular depth estimation by 30% on MSE using transfer learning with EfficientNetB3 and U-Net.

• Improved monocular depth estimation by 30% (MSE) through transfer learning with EfficientNetB3 + U-Net

MSc, AI & Data Science

Heinrich Heine University Düsseldorf

Data Engineering Specialization 2

EPAM School of Digital Engineering

BEng, Electrical and Computer Engineering

Nazarbayev University