Kirill Rybkin
Applied AI / ML Engineer · Founder
Over eight years in machine learning I have learned to build full-cycle LLM systems, and I mean the whole path: collecting and preparing the dataset, fine-tuning (RL methods included), quantization and self-hosted inference, then the backend, the frontend and the product analytics. A product mindset is what let me grow my own AI assistant, EssayAI, to 173,000 users in four months and cut costs several times over in other commercial projects.
rybkkir@gmail.com ·
github.com/Oxonomy ·
linkedin.com/in/kirill-rybkin-ai ·
huggingface.co/KirillR
Olympiads
- International Data Analysis Olympiad 2020: first place on the public leaderboard and second on the private one
- International Data Analysis Olympiad 2021: particle type and energy from detector images, ROC AUC 0.9993
Experience
- Founder & ML Engineer · EssayAI
- Machine Learning Engineer · ZakupkiAI
- Founder & Team Lead · DogBreeder
- Machine Learning Engineer · High Tech Engineering
Projects
- Open datasets: BearAtlas (2.2 million texts) and Infinity-Instruct-RU-Synthetic (6.16 million dialogues)
- ru-small-doge: a small Russian language model
- Ozon Seller Analytics: an ML stack for a marketplace seller
- RewriteRL: RL fine-tuning of a specialized model for a custom reward
- A pipeline for testing product hypotheses
- Anora: audience research for a psychological support service