Lyubomir Stoykov

Previously Staff Engineer at Yelp (full timeline)
Now: Building in the AI space

What I'm working on

Build

Hallvi: deploys and runs your apps on a server you own.

Content

YouTube videos, blog articles & ad-hoc thoughts on X.

Courses

AI Engineering: Agents, Workflows and Evals

Coming soon

For companies

AI agents and workflows

For the work your business wants to automate

Work with me →

Hi, I'm Hallvi.

The agent for self-hosted software.

Visit Hallvi →
Product architecture view showing an application's containers, private network, and persistent storage.
The Architecture view of a real Paperless-ngx deployment: reverse proxy, database, cache, and storage.

Latest videos

All videos

About me

Hello, I'm Lyubomir. I grew up in Bulgaria, left home at 14 for a math high school, and moved to Germany at 19. I studied Computer Science at TU Munich and graduated with M.Sc (Distinction). After graduating, I worked at a startup incubator in Munich, then joined Yelp in Hamburg, where I worked on Ads for six years and went from Software Engineer to Staff Software Engineer. I led projects on ads that reach millions of users, appear at the top of Yelp's most visited pages, and bring in tens of millions of dollars a year. I left my job in mid‑2025, moved back to Bulgaria, and started working independently in 2026.

Full timeline
  1. 2026 →

    Independent

    Building products, content, and selective consulting

  2. 2025 - 2026

    6-month Sabbatical

    Moved back to Bulgaria

  3. 2024 - 2025

    Staff Software Engineer

    Yelp · Ads · Hamburg · Munich

  4. 2021 - 2024

    Senior Software Engineer

    Yelp · Ads · Hamburg

  5. 2019 - 2021

    Software Engineer

    Yelp · Ads · Hamburg

  6. 2018 - 2019

    Software Engineer

    UnternehmerTUM · Digital Product School · Munich

  7. 2012 - 2018

    BSc & MSc, Computer Science (Distinction)

    TU Munich · Thesis presented at the Middleware Conference, Las Vegas

Latest writing

All posts
What's Left For Humans After AI?AI is taking over what humans used to do. What remains is vision and deep understanding. The Anatomy of an LLM RequestWhat actually gets sent to an LLM: system prompts, tools, MCP, memory, skills, messages, caching, and the complete tool-use loop. The Lifecycle of a Local AI Model: A Beginner's MapA beginner's map to Hugging Face, Transformers, PyTorch, GGUF, safetensors, Ollama, LM Studio, MLX, vLLM, and the rest of the vocabulary.