# Aleksei Bukhalov

> Aleksei Bukhalov is an SDET at MariaDB, working on distributed systems, performance, and test infrastructure. Outside work, he experiments with local LLMs and agent tooling.

## Professional focus

- Testing complex backend, database, and distributed systems
- Performance, failure modes, and production-like environments
- AWS, CI/CD, test infrastructure, and observability

## Selected builds

- [serbian-dub](https://github.com/liqlos/serbian-dub): A Russian-to-Serbian video dubbing pipeline with transcription, speaker-aware translation, voice synthesis, timing, mixing, and automated QA.
- [compact_llm_language](https://github.com/liqlos/compact_llm_language): Completed research on hidden-state recurrence in Qwen3.5-2B; no replicated reasoning gain in the tested recipes.
- [requirements-miner](https://github.com/liqlos/requirements-miner): Requirements tooling with isolated expert review and preserved dissent.

## Small iOS experiments

Aleksei released seven small iOS apps to learn the complete path from an idea through App Review and into the App Store. They are presented as learning experiments, not as claims of major product impact.

- [Inhale](https://bukhalov.com/inhaleapp/)
- [Lease Log](https://bukhalov.com/leaselog/)
- [Card Notebook](https://bukhalov.com/card-notebook/)
- [HOA Notebook](https://bukhalov.com/hoa-notebook/)
- [Mole Tracker](https://bukhalov.com/skinlog/)
- [MenoLog](https://bukhalov.com/menolog/)
- [Florida Bills](https://bukhalov.com/florida-bills/)
- [All iOS experiments](https://bukhalov.com/apps/)

## Other experiments

- [Hofstede Map](https://bukhalov.com/hofstmap/): An interactive web map, not an iOS app.

## Writing

- [Writing archive](https://bukhalov.com/writing/): Articles on local LLM inference and throughput, performance, and systems under load. Each published piece has a separate canonical HTML page.
- [How Local LLM Throughput Scales with Multiple Agents](https://bukhalov.com/writing/local-llm-throughput/): Published August 28, 2026. Why total output can rise even while each agent slows down.

## Links

- [Website](https://bukhalov.com/)
- [GitHub](https://github.com/liqlos)
- [LinkedIn](https://www.linkedin.com/in/abukhalov/)
- [Email](mailto:alex@bukhalov.com)