# Pavel Fedorov

> Personal portfolio of Pavel Fedorov, an ML systems engineer focused on LLM and speech inference optimization

The portfolio covers GPU performance, real-time serving, inference costs, work experience, education, and teaching. The HTML portfolio is the canonical source; the Markdown version is generated from its content.

## Portfolio

- [Portfolio](https://pavelfedorov.io/): Canonical website with Work, Projects, Blog, and About sections
- [Full portfolio in Markdown](https://pavelfedorov.io/index.md): Readable text including experience, projects, and contact links
- [Work](https://pavelfedorov.io/#work): Professional experience
- [Projects](https://pavelfedorov.io/#projects): Inference systems and optimization results
- [About](https://pavelfedorov.io/#about): Background, education, and teaching

## Engineering blog

- [All articles](https://pavelfedorov.io/blog/): LLM and speech inference worklogs
- [Nearly 6× faster with a compile cache. B200 workers in 45 seconds.](https://pavelfedorov.io/blog/vllm-compile-cache/): An S3 compile cache made H200 startup nearly 6× faster. Dynamo Snapshot brought B200 workers online in about 45 seconds.
- [The TTS bottleneck before the first sound](https://pavelfedorov.io/blog/tts-first-audio/): Reworking VITS from a model invocation into a compiled, batched pipeline for interactive voice calls.
- [Scaling Qwen voice serving under a one-second budget](https://pavelfedorov.io/blog/qwen-voice-capacity/): How I tuned graph coverage, speculative decoding, and precision for voice-agent traffic with a one-second LLM budget.