About

Sofia Iskandarova | Product Leader (AI & Developer Tools)

Author of the blog “Controlled Hallucinations”. I write about AI, large language models and the tools developers actually work with. I put models and services through real tasks.

Telegram: @sofya_iskandarova · Habr: controlled_hallucinations · LinkedIn: Sofia Iskandarova · Email: sofi@gotacat.dev

I lead the AI Products Development department at Rostelecom. The corporate AI platform is mine, along with everything that grew around it: agentic assistants, RAG-based search, computer vision and document processing, developer tooling. Before that I ran the same platform as its product owner, and earlier still I owned the chatbot and voice assistant at Auchan.

One moment interests me above the rest: when a language model stops being a demo and turns into part of an actual workflow. With an owner, metrics and a price tag. That is what I write about: agentic and multi-agent systems, RAG, developer tools. What survives real load, and what falls apart on first contact with production.

Outside product work I judge technology awards and publish research. Fellow of BCS, The Chartered Institute for IT; judge of the Globee Awards for Technology; member of the SKOLKOVO Global Innovator expert council; papers on multi-agent systems and enterprise AI. Whatever I pick up along the way goes into my Telegram channel “Controlled Hallucinations”.

Core Skills & Technologies

Product & Strategy

• Strategy and roadmap for an enterprise AI platform: from a single product to a suite of services.
• Taking internal AI services outside the company, to external organisations.
• Product research, prioritisation, unit economics, and the business case for engineering investment.

AI Systems

• Agentic and multi-agent architectures.
• Search and retrieval-augmented generation over enterprise data.
• Computer vision and document processing: recognition, data extraction, searchable archives.
• Assistants for developers: code generation, review, documentation.
• Choosing between fine-tuning, RAG, and prompting for a given task and budget.

Team & Adoption

• Leading a multi-team product, engineering, and Data Science organisation.
• The path from idea to operation: discovery, delivery, support, measuring impact.
• Education and advocacy: internal programmes, conferences, teaching.