Tural Mehtiyev

Data Scientist · CS PhD Student at NC State

I'm a data scientist with expertise in statistics, experimental design, and predictive modeling, paired with product-oriented thinking, research depth, and hands-on engineering skills. I am suited to Product Data Scientist, ML/AI Engineer, and Applied Scientist roles in software-intensive domains, where understanding both the data and the system that generates it is essential.

My research develops diagnostic frameworks for LLM agent reliability: failure detection, automated testing, and root cause analysis. I focus on understanding how AI coding agents behave in practice, using large scale behavioral trajectory analysis, causal inference, and natural experiments.

Previously, I spent over 6 years as a product data scientist, building AI powered products, designing A/B tests, and developing predictive models that drove product decisions.

Focus areas: LLM agent behavior, software engineering automation, empirical methods, product experimentation, A/B testing, causal analysis.

GitHub LinkedIn Google Scholar
Tural Mehtiyev

Expertise

Technical Skills

💻 Programming & Data Wrangling

Python SQL Excel SAS Spark Hadoop

📊 Statistics & Experimentation

A/B Testing Hypothesis Testing EDA Inferential Statistics

🤖 Machine Learning & AI

PyTorch Scikit-learn Deep Learning Computer Vision NLP LLM Agents LangChain RAG MCP MLOps

☁ Cloud & Tools

GCP BigQuery Vertex AI Airflow Tableau Power BI Git Docker Linux

Work

Featured Projects

testflow-ai

↗

AI-powered black-box end-to-end testing SaaS platform. Automates test generation and execution using intelligent agents.

AI SaaS Testing Python

EEG-Gaze-Prediction

↗

EEG-DCViT: Novel deep learning model combining CNNs with Vision Transformers, achieving state-of-the-art results in EEG-based gaze prediction.

Deep Learning Vision Transformer PyTorch

MPI-Shared-Memory-Programming

↗

Parallel Sparse Matrix-Vector Multiplication using MPI, OpenMP, and CUDA for high-performance distributed computing.

MPI OpenMP CUDA C

stencil-optimization-simd-omp

↗

Achieved 29.5x speedup on stencil computations through SIMD vectorization and OpenMP parallelization optimizations.

SIMD OpenMP Optimization C

llm-api-tm

↗

LLM-based dynamic API function calling evaluation framework. Benchmarks how well language models handle tool-use and API integration.

LLM API Evaluation Python

SUMMA Matrix Multiplication

↗

Distributed dense matrix-matrix multiplication using the SUMMA algorithm with MPI for scalable parallel linear algebra.

SUMMA MPI Linear Algebra C

Career

Work Experience

Aug 2024 — Present

Doctoral Researcher — Testing Agentic AI Systems

North Carolina State University · Full-time · Raleigh, NC

  • Agent Evaluation Framework: Developing systematic benchmarking methodologies for diagnosing failures in LLM-powered agents using behavioral trajectory analysis and pattern detection
  • Early Failure Detection: Building diagnostic tools to identify failure patterns during agent execution for resource-efficient intervention in software development workflows
  • Cross-Framework Analysis: Investigating generalization of failure patterns across multiple agent architectures and model scales to identify universal diagnostic signals
Jan 2026 — Jun 2026

Graduate Researcher — AI Governance & Policy

State of North Carolina · Part-time · Raleigh, NC

  • Conducted technical safety research for the AI Governance & Safety subcommittee at North Carolina AI Leadership Council, researching comprehensive assessments of AI system testing, validation, standards, monitoring, traceability, and auditing frameworks to inform North Carolina's statewide AI policy and legislative recommendations
  • Analyzed state capabilities and gaps in pre-deployment safety testing, bias auditing, red-teaming protocols, and incident response mechanisms for high-risk AI applications with findings informing recommendations to the Governor's office
  • Evaluated emerging challenges in agentic AI systems and algorithmic evaluation methodologies, synthesizing national technical frameworks with North Carolina's policy needs and implementation considerations
Aug 2022 — Jul 2024

Data Scientist

Azericard · Full-time · Baku, AZ

  • Internal Knowledge Assistant (LLM/RAG): Built an LLM-powered internal chatbot to enable stakeholders to query comprehensive card product information from the internal knowledge base
  • Churn Prediction: Developed predictive models using ensemble methods to identify causes of customer attrition, reducing churn in targeted retention campaigns
Oct 2020 — Jul 2022

Data Analyst

Kapital Bank OJSC · Full-time · Baku, AZ

  • Automated Reporting Pipeline: Collaborated with stakeholders to formulate data solutions using SQL and Tableau, reducing report generation from days to hours
  • Product Launch Analytics: Analyzed performance metrics and product telemetry for newly shipped card products, conducting A/B tests
Jul 2020 — Jun 2022

Data Analytics Consultant

Nexia · Contract · Baku, AZ

  • Analyzed large datasets to uncover trends, insights, and business opportunities for clients in the transportation and energy sectors
Apr 2019 — Sep 2020

Junior Data Scientist

Azerconnect LLC · Full-time · Baku, AZ

  • Marketing Campaign Optimization: Built customer micro-segmentation models, designed statistical experiments with hypothesis testing on campaigns
  • 4G Network Optimization: Utilized time series models and clustering algorithms on large-scale historical data to forecast network demand

Research

Publications

Cited by 38 · h-index 2 · Google Scholar

Beyond Resolution Rates: Behavioral Drivers of Coding Agent Success and Failure

T Mehtiyev, W Assunção

arXiv preprint arXiv:2604.02547, 2026

Advancing EEG-Based Gaze Prediction Using Depthwise Separable Convolution and Enhanced Pre-Processing

ML Key, T Mehtiyev, X Qu

International Conference on Human-Computer Interaction, 2024

Advancing EEG-Based Gaze Prediction: Depthwise Separable Convolution and Pre-Processing Enhancements in EEGViT

ML Key, T Mehtiyev, X Qu

Preprint

View Google Scholar Profile →

Background

Education

Aug 2024 — Expected Dec 2027

PhD, Computer Science

North Carolina State University · Raleigh, NC

GPA: 4.00 / 4.00

Focus: Failure diagnosis and behavioral analysis of LLM-powered coding agents

Aug 2022 — Jun 2024

Master of Science, Computer Science (Online)

The George Washington University · Washington, DC

GPA: 3.93 / 4.00

Specialization: Machine Learning

Sep 2014 — Jun 2019

Bachelor of Business Administration

ADA University · Baku, AZ

GPA: 3.88 / 4.00

Focus: Business Analytics and Statistics

Communication

Languages

English

Professional Proficiency

Turkish

Professional Proficiency

Japanese

Elementary Proficiency

Russian

Limited Working Proficiency

Get in Touch

Let's Connect

Interested in discussing research or have questions? Feel free to reach out.

tmehtiy@ncsu.edu