@zeulinn
This developer is so mysterious, even their code has commitment issues.
An end-to-end AI pipeline that watches gameplay footage and generates a synthetic streamer avatar that reacts, trained on real YouTube streamers playing Signalis.
View ProjectAnother version of the project. Slightly different approach from the one we submitted as a team. The idea is to predict facial landmarks and synthesize an avatar based on them. The project does not actually use prediction for now, as the pipeline for data preparation needs to be change (it seems that there is a ground-truth data leakage into train and test splits). Attached is the video with the idea. NB: It's a side submission, just wanted the effort to be visible
View ProjectAn autonomous pipeline that takes a company's existing content, a content goal, and optional competitor data, then classifies the query intent, extracts the brand voice via RAG, and infers the competitive landscape. It synthesizes these into a content brief, writes a structured draft, injects GEO signals proven to increase AI engine citation rates, and runs it through a multi-judge quality loop — automatically rewriting until the content meets the quality threshold. The final output is culturally localized into multiple languages, publication-ready, and scored across four dimensions: customer fit, GEO compliance, competitor differentiation, and human naturalness.
View ProjectSkills include: Turning coffee into code, debugging by staring intensely at the screen, and mastering the art of Stack Overflow copy-paste.
Social links? Pfft. I communicate exclusively via binary smoke signals.
