Aruzhan Nurmanova

AI/ML Engineer

Dreamlab Technologies

Built production AI systems for legal documents, chatbots, model serving, and automated code audits, with reported reductions in GPU memory use and inference costs.

• Deployed 8-bit quantized Gemma 4 26B via vLLM, cutting GPU memory usage by 50% and boosting inference throughput by 15%. • Cut inference costs by 40% by routing simple queries to smaller and faster inference models. • Engineered a production multilingual OCR pipeline for scanned Russian and Kazakh legal PDFs, combining Dots-OCR-1.5, OpenCV-based table detection, and NER/LLM-based extraction into structured JSON output.

Data Scientist, Research Assistant

Institute of Smart Systems and Artificial Intelligence (ISSAI)

Researched hydrological forecasting and built data and OCR pipelines, including rainfall-runoff models for 85 basins with an NSE of 0.93.

• Trained LSTM-based models for rainfall-runoff prediction across 85 Kazakhstani basins, achieving NSE of 0.93. • Built an automated data retrieval system in Google Earth Engine for climate and hydrological variables, and applied QGIS for volumetric basin evaluation and geospatial visualization. • Experimented with multiple architectures (LSTM, EA-LSTM, Transformer, Mamba) for hydrological forecasting. • Implemented a custom parsing and OCR system to digitize 1,000+ scanned files into structured datasets.

Mentor

Technovation Girls

Mentored a team that reached the Technovation World Summit finals and won the Technical Prize for an AI app supporting children with cochlear implants.

• Led a team to the finals of Technovation World Summit 2024 at NVIDIA HQ (top 0.2% among 11,000+ projects). • Won “Technical Prize” for developing an AI app that supports children with cochlear implants.

Master of Science, Data Science

Nazarbayev University

Bachelor of Science, Computer Science

Nazarbayev University