Skip to content

Sanzhar Soltybekov

@sanzhar-soltybekovConnected with DSML. This person's Telegram account has been confirmed through DSML Kazakhstan.

AI Engineer and Machine Learning Researcher

I’ve been building AI tools for career coaching and experimenting with machine learning for drug discovery. Lately, I’ve been comparing ways to pick promising compounds, and I enjoy making the results useful in real systems.

Open to

Learning with othersJoining competitions & hackathonsPaper discussionsResearch collaborationCommunity initiatives

Career

Experience: 0 years

Experience

AI Engineer

Mentor

Sep 2026 — Present

Builds AI career coaching features and backend services, including user memory, multilingual profile reviews, LLM evaluation, and personalized practice.

• Implemented Anthropic-inspired Dreaming for an AI career coach. A background worker extracts and deduplicates project facts from completed roadmap steps, storing source quotes and message IDs for reuse across each user’s career plans. • Built an authenticated Go API and Python/FastAPI pipeline for AI career-profile reviews in Russian and Kazakh, with structured OpenAI outputs, rate limits, and PostgreSQL persistence. • Created a paired LLM evaluation harness to compare draft and approved mentor guidance on 8 synthetic profiles, using the same prompt and OpenAI client as the API.

Research Assistant – Machine Learning for Drug Discovery

Nazarbayev University

A research university in Astana, Kazakhstan, with programs across science, engineering, and other fields.

Aug 2026 — Present

Applies machine learning to drug discovery, comparing compound selection methods and building affinity prediction models for computational screening.

• Adapted DeepDTA in PyTorch to compare MSE and SPO+ for compound selection. Trained both variants on 24,358 molecule–protein pairs and evaluated on 3,697 pairs with held-out molecular scaffolds. • Observed 7.0% lower mean Regret@1 with SPO+ (1.408 to 1.310 pKd) across 194 validation proteins in a single-seed pilot. Measured the trade-off in prediction accuracy and estimated uncertainty with a paired bootstrap by protein.

AI Development Intern

Business and Technology Services (BTS)

May 2026 — Aug 2026

Built AI automation and messaging integrations at BTS, supporting over 1,000 employees and cutting WhatsApp message detection delay by 80%.

• Developed an AI-powered browser automation agent based on Computer Use (Vision-to-Action) architecture, without relying on Playwright/CDP dependencies. • Built and deployed an asynchronous FastAPI gateway integrating Telegram and WhatsApp with the corporate FAIR AI platform, supporting an active user base of 1,000+ employees. • Reduced WhatsApp inbound-message detection delay from 5 seconds to 1 second (80% lower) by implementing tuned Twilio API polling in a private Kubernetes network without public webhooks.

Education

Bachelor of Science, Computer Science

Nazarbayev University

2025 — 2029

Peer-to-Peer Software Engineering Program

Alem School

2025 — 2025

Skills

PythonGoSQLGitPostmanGitLabAI AgentsComputer Use (GUI Automation)RAGFunction CallingContext Window ManagementPyTorchscikit-learnRDKitpandasNumPyHugging FaceFastAPIPostgreSQLRedis