CoAT: Corpus of Artificial Texts
A large-scale Russian-language dataset containing both human-written and AI-generated texts.
Cambridge
About
Marat Saidov
Machine Learning Engineer
I am a versatile and dedicated engineer with a strong focus on ML product development for natural languages and human speech.
Work
I work at Applied Sciences Group, Microsoft. As part of this group, I contribute to the LLM platform and Text Intelligence scenarios, enabling language models across Microsoft products, including Windows OS.
Publications
Follow the full publication trail on Google Scholar.
LUNA: A Framework for Language Understanding and Naturalness Assessment
A unified interface for 20 NLG evaluation metrics, designed for comprehensive generated-text evaluation.
arXivFindings of the RuATD Shared Task 2022
Results from the Dialogue Evaluation shared task on artificial text detection in Russian.
arXivProjects
Artificial Text Detection
My BSc thesis project studying how language models distinguish artificial texts from human-written ones, connected with my research at CPLab.
GitHubJob Listing Matcher
A chat-based job-matching assistant. The user describes their experience and what they are looking for in plain language; the system turns that into a structured profile, retrieves the best-matching vacancies from a vector database, re-ranks them with a trained model, and presents them next to the conversation, which the user can keep refining to request more results.
GitHubOpenClaw Azure self-host template
A sanitized template for running a personalized OpenClaw agent on a small Azure Linux VM, with GitHub Copilot-backed model access, pull-based GitOps, scheduled workflows, cloud-drive guardrails, Azure Speech integration, and disposable Crabbox/Daytona dev boxes for heavier software testing.
GitHub Write-up