Бекнұр Қалмаханбет

AI Research Intern

Inception (G42)

AI company within Abu Dhabi-based technology group G42.

Built a lithology classifier across 223 wells, reaching 0.70 macro F1 on the FORCE 2020 benchmark and calibrating predicted probabilities.

• Implemented a Transformer-based lithology classifier on LAS well-log data covering 223 wells and 10–12 lithofacies classes. • Achieved a macro F1 of 0.70 on the FORCE 2020 benchmark. • Applied post-hoc calibration via decision-threshold tuning to improve reliability of predicted class probabilities.

MLOps Specialist (Computer Vision)

Cleverest Technologies

Built and operated a multi-camera analytics platform across three sites, deploying four detection models in air-gapped environments.

• Built and shipped UMAI, a modular video analytics platform processing 4 concurrent camera streams at 25 FPS across 3 deployment sites. • Trained and deployed four detection models — fight detection, abandoned object, smoke, and automobile plate recognition — holding consistent real-world performance across varying lighting and traffic conditions. • Deployed and maintained a production ML system in air-gapped, on-premise environments for the Department of Presidential Affairs and the Ministry of Defense, running unattended for 2 months.

Research Assistant

Nazarbayev University

Research university in Kazakhstan.

Automated GPU library benchmarking across 130 configurations and traced performance gaps of up to 19% to memory and thread metrics.

• Automated a micro-benchmark testing five CNN libraries (cuDNN, cuBLAS, TensorRT, Darknet, direct convolution) across 130 configurations, auto-parsing results into a reusable dataset. • Identified performance gaps of up to 19% between GEMM libraries and traced them to memory- and thread-level metrics using nvprof.

MSc, Computer Vision

Mohamed bin Zayed University of Artificial Intelligence

BSc, Computer Science

Nazarbayev University