Tim. Prague.
Machine learning and software.
I'm 16. I publish datasets, compete on Kaggle, build products, and occasionally make WhatsApp run on a BlackBerry.
// 01 Products
currently buildingVirlya
Turns a brief into research, copy, visuals, reels, a publishing plan, and an approval workflow.
ForkIt
Swipe-based GitHub discovery with real starring and preference signals collected for future ranking.
// 02 Hackathons + competitions
team buildsPronea Hackathon: spndr
An AI CFO prototype built and shipped at the Pronea Hackathon in Partizánske, Slovakia.
AI Olympiad: VoltPlan
EV demand prediction and charging-infrastructure planning, built with team notokens.
// 03 Data + ML
public workBTC News Reactions
Multilingual crypto-news observations joined to Bitcoin price reactions.
Kaggle S6E8
Stacking pipeline with negative results and the final selection mistake documented.
// Open source sandbox // utilities
claude-code-toolbelt
Five Claude Code plugins for multi-model review, structured ideation, API discovery, supply-chain inspection, and guardrails.
bb10-whatsapp
A native WhatsApp client for BlackBerry 10 and legacy Android, backed by Evolution API and a small Node service.
claude-code-earcon
Zero-runtime-dependency sound notifications for permission prompts and completed Claude Code turns.
// Working rules // what survives contact with the data
Start with a baseline
A score is meaningless until it beats something simple. Every model gets compared before it gets described.
Keep the holdout clean
Leakage is a bug, not a leaderboard strategy. Evaluation mistakes belong in the post-mortem with the wins.
Publish the artifact
Code, data cards, provenance, limitations, and negative results make a project useful beyond the final screenshot.
Build the whole loop
Data and models matter, but so do the interface, infrastructure, feedback events, and the person using the result.
// Engineer credentials & profiles
TIMOFEJ GOLOBOKOV // BIO
I'm a 16-year-old SSPŠ student in Prague working across applied machine learning, public datasets, and product engineering. Virlya and ForkIt are the two active products.
I compete on Kaggle, publish data on Hugging Face, and keep the evaluation notes next to the result. Outside ML, I build developer tools and software for constraints that are fun to take seriously, including BlackBerry 10.