Marat Gazizulin — ML Engineer
Agents, RAG and multimodal pipelines — from architecture to production.
Requirements become components, data flows and task distribution.
Text, image, video, audio and 3D — company GPUs and external APIs behind one platform.
Dataset-based evaluation, tracing and monitoring before anything ships.
Start with the problem and its constraints. An architecture decision should explain why an option fits and how to test it in practice.
ML Engineer → Lead ML Engineer → CTO
A platform for generating and editing text, image, video, audio and 3D. Built from scratch and launched.
ML Engineer
An LLM game master for role-playing games in Telegram and Discord.
Lead ML Engineer
Requirements, architecture and environments on Runpod and ComfyUI: custom nodes, character consistency, pose control and LoRA training.
Work in the open: local generative studios on my own 16 GB consumer GPU, and an experiment in controlling what LLM agents may disclose.
A self-contained local studio: projects, image generation and editing, image- or text-to-3D, video with sound, and model management with checksum verification.
Repository
“Change the red teapot to deep blue ceramic. Keep the table and window light.”
A local studio and MCP server for HiDream-O1: React interface, Flask API with a bounded GPU queue, SSE progress and tools for agents.
RepositoryPolicy-controlled disclosure for LLM agents
An experimental Python SDK: the model sees opaque references, and original values reach only authorized tool fields after a fresh permission check. I compared it with a handwritten boundary on two synthetic agent apps.
Key finding: data identity and disclosure scope are different things. A contact shared by two issues was safely rejected: a false denial, not a leak.
Experimental source, not a production security product
April 2026 — presentNow
ML Engineer
Technical implementation of AI tools for business process automation at a company of 250 people.
January 2024 — March 2026
ML Engineer → Lead ML Engineer → CTO
Built and launched Artillect from scratch, supporting text, image, video, audio and 3D generation and editing. Designed the architecture, backend APIs, task distribution and render workers; integrated models on a company GPU server through ComfyUI with external APIs through fal.ai. Implemented RAG for prompt generation based on the selected prompt type. Later owned technical planning, hiring, team coordination and release support.
February — August 2025
ML Engineer
Part-time project engagement. Built an AI game master for Telegram and Discord RPGs: multiple Dify workflows with over 100 nodes in total, different models, prompts and memory. Set up automated dataset-based answer evaluation, tracing and monitoring. Developed Python APIs, tests, MySQL integration, usage limits and subscriptions.
February — June 2025
Lead ML Engineer
Part-time project engagement. Clarified requirements and designed fashion-content generation pipelines using Runpod and ComfyUI, including custom nodes, character consistency, pose control and LoRA training.
May 2021 — June 2023
Software Engineer · C# / Unity
Built game and interactive systems: gameplay, multiplayer, tooling and backend integrations. The software engineering foundation behind my AI work.
From requirements to architecture: components, data flows, task distribution and model integration.
Multi-step Dify workflows with multiple models and memory. RAG for scenario-specific prompt generation.
Automated dataset-based answer evaluation, quality and performance analysis. Tracing and monitoring, including Langfuse.
Python APIs, databases and tests. External model integration and execution on company-hosted GPUs.
Image, video, audio and 3D generation and editing. ComfyUI, custom nodes and LoRA training for visual tasks.
05 — Contact
Interested in ML Engineer roles involving AI systems, agents, RAG and automation, with responsibility for architecture, quality evaluation and delivery.
