National legal AI search engine for Azerbaijan. I designed and led the ML architecture end to end: document processing and chunking for hierarchical legal texts, hybrid retrieval that combines lexical and semantic methods, learning-to-rank, re-ranking, and quantized LLM inference on H100/H200 GPUs. Built with a team I trained from scratch.
Work
Things I designed and built, mostly between 2023 and 2026. All of them now run inside the institutions they were built for.
A tuition-free program that trains AI engineers from zero. With a colleague I designed the curriculum, the Mathematics for AI and AI Engineering tracks. I built the admissions process and the LMS structure, and interviewed faculty candidates.
Interactive AI lessons for schoolkids, in Azerbaijani, in the spirit of Brilliant and Elements of AI. I designed the course structure and the lesson content with the team I trained.
A Kaggle analogue for Azerbaijan: companies open their real datasets and problems to the local ML community, and solutions compete on objective metrics. I designed the platform concept and the competition format, again with my team.
Azerbaijan’s first AI supercomputing center: NVIDIA H200 GPUs on Lenovo ThinkSystem, InfiniBand NDR, DDN storage. I contributed to the technical design and specification, and to project management, validation and end-to-end delivery.
The AI Team
Alongside these projects I built and trained an AI engineering team, from zero to eighteen people, starting with interns. They are now in graduate programs at EPFL, TUM, The University of Edinburgh, Johns Hopkins University, The University of Manchester, and Imperial College London.
National AI Strategy
Azerbaijan’s national plan for AI. I took direct part in preparing it, and everything above, from the search engine to the team, was part of its execution.
Earlier
Before ML I spent a decade in telecom: building the products, running the infrastructure, leading the teams. At Azercell I built and ran the products and services a ten-million-subscriber network sells, along with the charging and provisioning systems underneath them. At TotalIT I designed network, system, data and distributed architectures, end to end, for the companies we served. Along the way I implemented things from their primary sources to understand them: the Diameter protocol from RFC 6733, the Raft consensus algorithm from the original paper, a WebAssembly compiler from the core spec.