- 🎓 Still in high school · already shipping code
- 💼 Backend developer · ML researcher
- 🌱 Currently learning NLP algorithms && agent security
- 📝 Six months of work at ECM Consulting as a test automation engineer
- 🏅 Kaggle Contributor
- 🏆 Prize-winner / absolute winner of national olympiads · finalist of international olympiads — full breakdown in the portfolio
Hey! I'm Roma — still in high school, but that doesn't stop me from building pet projects, taking outsource work, and winning olympiads along the way.
I've been writing Python for 4+ years: started with Telegram bots, moved on to APIs and backend development, and for the past 2 years I've been deep in AI/ML — CV and agents are my current obsessions. Also growing my Kaggle profile bit by bit.
Had a stint as a Test Automation Engineer at ECM-Consulting (Prometheus + Grafana) — great experience working on real infra. Now I'm back to competing, building, and learning. If you want to collaborate — you're welcome to join)
| Project | Stack | Description | Links |
|---|---|---|---|
| Motivation bot | Python Telebot Ollama |
Parse events from google calendar and gives you personal motivation messages based on LLM every day | Code |
Boilerplate / starter repos for quick project bootstrap
| Project | Year | Stack | Description | Links |
|---|---|---|---|---|
| KotlinFish | 2026 | Kotlin Spring boot |
Backend starter template in Kotlin (CRUD + JWT Auth) | Code |
| FastAPI-Fish | 2026 | Python FastAPI |
Backend starter template with FastAPI (CRUD) | Code |
Projects built while studying — courses, tutorials, self-driven experiments
| Project | Year | Direction | Course / Source | Description | Links |
|---|---|---|---|---|---|
| Deepfake classifier | 2025 | ML: Computer Vision | Yandex Lyceum | NN based on ResNet50 that can distinguish generated faces from real ones. F1 score abt 0.888 | Notebook |
| Low-resource translation | 2026 | ML: Machine translation | DATA DOJO | Fine-tune NLLB-200 0.6B on 0.5M texts parallel corpus to translate sentences of 5-20 words. Result: BLEU 13.9 — top 23 of 400+ members | Code · Article (RU) |
| Objects counter | 2026 | ML: Computer vision | HSE&MWS SSDA | Using regression instead of classification gives accuracy 71.743 and absolute 1st place | Notebook |
Machine Learning
Computer vision · RecSys · Deep learning · Fine-tuning
Backend
APIs · relational data · caching · services
Infrastructure & Tools
Monitoring · containers · version control · API testing