ML Engineer II, Copilot Studio
Microsoft
Microsoft is a global technology company whose products include Azure, Windows, Microsoft 365, and Copilot.
Nov 2025 — PresentTrains and deploys small language models for agent evaluation and sentiment classification, with a reported win over GPT5-mini on a holdout test set.
• Trained analytical SLM that judges LLM agent decisions (SFT, GRPO, Reinforce, SAPO, SDPO). The model outperformed GPT5-mini on holdout testset
• Trained and deployed (AML) latent reasoning intensive sentiment classification SLM (mistral-7B, qwen3.5) using RL techniques for Online Metrics team - microsoft/Sentiment-Analysis-7B
• Contributed to ThinkingBox, LLM agent evaluation framework - code
• Hackathon project: Built Coding Diffusion language model using message-level reward RL that showed 71.3% score on SWE-Bench_Verified
Lead Researcher, Founder
AILabs Technologies, Inc.
5-employee R&D company
Feb 2016 — Oct 2025Founded and led a five-person R&D team, building LLM, RAG, and speech tools, plus customer support software that generated over $200,000 in revenue.
• Built oLLM, LLM inference library that allows running large models on consumer GPUs (~2.8k on Github) - code
• Built Webapi.ai, first AI agents builder that leverages GPT-3 few-short learning capability (this concept is used in thousands of products like Microsoft Copilot Studio, Intercom, etc)
• Built peftee, efficient LLM fine-tuning library for consumer GPUs - code
• Introduced Contextual chunking with keywords, a method that improves RAG accuracy (this concept is used by many companies like Anthropic) - article | meta/llama-cookbook
Junior Researcher
Nazarbayev University
Nazarbayev University is a research university in Astana, Kazakhstan, with academic programs in science, engineering, and other fields.
Built search, annotation, and text normalization tools for Kazakh language data, and contributed to five research papers on the corpus project.
• Developed a Search Engine, including 4-word autocomplete using a large Kazakh language corpus text.
• Developed a text annotation tool (POS, Morphology, Syntax tagging) that allowed for annotating a large amount of Kazakh text and creating an annotated Kazakh language corpus as a result
• Developed a text-normalization tool using Java for the Kazakh language
• Developed tabular data association/relation algorithm using C++ that resulted in multiple Research papers
• Contributed to 5 Research papers related to the Kazakh corpus project, including the Search engine and annotation tool
