Marsel Sultanov

Stealth-mode personal R&D initiative (Delaware LLC) bridging applied GenAI engineering with foundational AI alignment research. Serves as a living portfolio and preparation ground for graduate studies in Europe. Through hands-on prototyping—using agentic architectures (LangGraph, MCP), retrieval-augmented systems, and AI-assisted development (Qwen/Qoder)—I explore autonomous multi-agent behavior, interpretability/control beyond RLHF, and lightweight, verifiable human-aligned AI in real-world contexts.

Led R&D and production deployment of GenAI systems to automate core business processes in a multinational FMCG environment with 10k+ employees. Key achievements include: integrating Anthropic MCP to connect LLMs with Google Drive, Slack, and codebases via Replit (5 scenarios tested, 1 deployed); reducing unstructured data analysis time by 60-70% via custom RAG pipelines (LangChain/LangGraph + FAISS) for PDFs, Excel, and internal docs; automating ERP material creation (text, screenshots, voice) to accelerate onboarding by ~30%; automating routine AI solution launches via BAT scripts; ensuring resilient AI access across 6 platforms via no-code proxies in restricted infra. Also adapted open-source multi-agent frameworks for internal automation and led cross-functional AI tool training.

Led evaluation and practical integration of emerging generative AI technologies into back-office workflows. Focused on training and deployment of AI tools (text-to-speech, CGI, music) and prototyped generative CGI, AI music platforms, digital avatars, and metaverse-related DAO interactions to inform future production pipelines. Trained localization teams on AI-enabled workflows to support multi-market adaptation.