BrandBlaze
Hackathon · Generative media
Data Scientist & AI Engineer, and an MSc Data Science student at Óbuda University in Budapest. I build machine learning systems, data pipelines, and agentic AI tools, with experience across banking, analytics, and e-commerce.
My work spans fraud detection, NLP assistants, large-scale analytics, and AI-powered production workflows. I also perform research on different AI & Data Science Topics.
Career
AI Production Engineer · ContractSpice Media Production, Budapest : Agentic AI for e-commerce content, production, and post-production workflows.
Data ScientistKapital Bank, Baku : Fraud detection, cash-replenishment pipelines for 1,000+ ATMs, and an NLP customer-service assistant.
Data ScientistBIRainy, Baku : Python and SQL analytics, Power BI reporting, and OCR document pipelines.
Education
MSc in Data ScienceÓbuda University, Budapest : Machine learning, statistical analysis, data pipelines, and reproducible research. TDK paper on benchmarking open-source LLMs for Azerbaijani, Fall 2025.
BSc in Information TechnologyAzerbaijan State Oil and Industry University : Google DSC, Innovation Club, and Debate Club; workshops, hackathons, and interdisciplinary teams.
Based in Budapest, Hungary.
The challenge
A product campaign tool built for the Backblaze Generative Media Hackathon. One reference photograph becomes campaign variants for different markets, channels and environments, with the product’s identity treated as a constraint.
Backblaze Generative Media Hackathon
2026
What we built
Ordinary image generation can change product geometry, materials, colors and markings. The challenge was to preserve those details while varying commercial art direction, and make each generated asset traceable to its source and review history.
My contribution
Claude analyzes the source into an identity contract and creates variant-specific art direction. Genblaze orchestrates reference-guided generation through GMI Cloud. Visual checks score each result; failed constraints trigger a targeted repair attempt. Backblaze B2 stores assets, manifests and lineage.
Outcome & demo
A pre-1.0 hackathon prototype with campaign review, approve/reject controls and an asset archive. Results that fail visual checks remain flagged rather than being presented as verified. Model-based QA supports human review; it does not guarantee product fidelity.
