Nursultan Asilbekov

Education

Conducted research on Mechanistic Interpretability of transformer-based AI models solving quadratic equations.
Applied interpretability techniques (logit lens, linear probin, etc.) and performed comparative studies of different model
architectures (standard vs. constant attention) to uncover their algorithmic behavior.
Presented the research poster at symposium.

Conducted hands-on workshops to help students understand backend development, covering topics like APIs, databases,
and authentication
Provided one-on-one mentoring to students, helping them debug and optimize their code.

Organizing and leading biweekly workshops on Neural Networks, Deep Learning, Transformers, and LLMs for
students. Designed hands-on, interactive sessions using Google Colab notebooks and illustrated slides to make
complex topics accessible.
Leading a team of 6 students developing an AI-powered course scheduler using Retrieval-Augmented Generation
(RAG). Mentoring members on data parsing, building vector databases, and integrating open-source LLMs, with
plans to develop a custom model tailored to our use case.

Developed a user-friendly meeting scheduling web application with enhanced optimization in a semester-long
“Decision Lab” program
Applied optimization methods using Gurobi and Google OR-Tools.
Built the backend with Django and the frontend with React.

Advanced through all three competition rounds, ranking in the top 0.4% out of 2,500 global participants
Developed an eco rideshare app with an intuitive UI and real-time route optimization algorithms to reduce carbon
emissions and promote sustainable travel.

Developed a user-friendly website for reviews to strengthen connections between restaurants and clients.
Designed and implemented full-stack features, improving accessibility and user engagement.