Alisher-Polat Kurmanayev

@myakkyyyConnected with DSML

ML Engineer

Machine Learning Engineer focused on recommender systems, applied machine learning, and production AI infrastructure.

Open to

Helping organize meetupsJoining pet projectsStartup collaborationSpeaking & workshopsProject collaboration

About

Currently building a next-generation music streaming platform at Freedom Music, designing recommendation systems and ML infrastructure from the ground up. Working across the full lifecycle of machine learning products, from data collection and feature engineering to retrieval, ranking, experimentation, and large-scale serving. Experience includes: * Designing recommendation systems using multi-stage Retrieval and Ranking architectures * Building vector-search pipelines with embedding-based retrieval and candidate generation * Developing ranking models for personalized content recommendations * Designing event-tracking and user analytics systems for ML datasets * Solving cold-start problems for users and content in large-scale recommendation platforms * Building FastAPI microservices and containerized ML infrastructure * Developing LLM applications, multi-agent systems, RAG pipelines, and workflow automation * Fine-tuning and deploying open-source language models Interested in: * Recommender Systems * Information Retrieval * Ranking and Personalization * Applied Machine Learning * ML Infrastructure and MLOps * Agentic AI Systems * Large-Scale Retrieval Architectures * Experimentation and A/B Testing Tech stack: Python, FastAPI, PyTorch, CatBoost, LightGBM, FAISS, Qdrant, PostgreSQL, Redis, Docker, Linux, LangGraph, LangChain, vLLM, scikit-learn, Prometheus, Grafana, Langfuse.

Career

Experience: 1 years

Experience

Machine Learning Engineer

Freedom Music

Jun 2026 — Present

Working on a new music-streaming recommendation system, including retrieval, ranking, event tracking, and cold-start approaches.

• Working on the design and development of a recommendation system for a greenfield music-streaming platform. • Contributing to a two-stage recommendation architecture with candidate retrieval using vector search and personalized ranking models. • Supporting the design of event-tracking pipelines for user interactions, including clicks, skips, listening sessions, and engagement signals. • Working on cold-start strategies for new users and tracks, ML infrastructure requirements, and experimentation workflows.

Machine Learning Engineer

TSARKA

Jan 2026 — May 2026

Built and evaluated LLM workflows and multi-agent tooling, including inference monitoring, tracing, preprocessing, and clustering exploration.

• Developed internal ML workflows using vLLM, FastAPI, and context retrieval to generate neutral client-facing case descriptions from structured operational reports. • Contributed to a LangGraph-based multi-agent framework with E2B sandboxing for controlled code execution and tool-based agent interactions. • Built token-efficient preprocessing and context-formatting components for large JSON payloads, improving model-readiness of noisy operational inputs.

Machine Learning Engineer

WONK

Apr 2026 — Jun 2026

Developed agent tools, context management, and personalization for an edtech platform that generates lesson plans.

• Contributed to a multi-agent edtech platform using LangGraph, LangChain, and LangMem to support automated lesson-plan generation for teachers. • Developed agent tools and FastAPI endpoints for integrating platform workflows with backend services. • Worked on context-management systems, including short-term session context and long-term memory, to improve continuity and relevance across agent interactions. • Implemented personalization features that adapt generated educational content to teacher preferences, classroom context, and lesson requirements.

Education

Bachelor of Science, Computer Science

Nazarbayev University

2022 — 2026

Selected projects

NeuraVA: Fine-Tuned LLM for Anonymous Student-Post Generation

Fine-tuned Qwen2.5-1.5B with QLoRA on anonymized Russian student-community posts to mimic short-text writing style. Built a Python/Pandas pipeline for extraction, cleaning, anonymization, JSONL conversion, supervised fine-tuning, and local inference; developed a Telegram bot for Turing-style guessing rounds with real and model-generated posts, scoring, and leaderboards.

Skills

PythonMLvLLMLangGraphPyTorch