Publication
Geometric Log-Space Ensembling for MAPE-Optimal Retail Revenue Forecasting: A Large-Scale Case Study
Aug 4, 2026 · International Journal of Computer Information Systems and Industrial Management Applications
OpenAlex citations: 0Open
Public profile
Daniel Dragonevskiy
@ispromashka
ML Engineer · AIRA
LLM/ML researcher. I aim to enroll in a top university for a master program and am looking to connect with researchers/professors and students from MBZUAI, MIT, Harvard, and Stanford for joint projects
MuscatNational Research University Higher School of Economics
Featured
A few things that best show this member's work and contribution.
How I can help
I am looking for colleagues for joint research (LLM, CV, optimization and ML) - including people from MBZUAI, MIT, Harvard, and Stanford - for collaborative projects and research, as well as the opportunity to pursue a master degree at these universities in the future
Looking to connect
What I can offer: help with collaborative research, conference talks, and journal publications, team-lead experience across 10+ hackathons and open source projects (TurboCat, intelligent agent systems for information search and analysis)
Research and publications
A compact preview of this member's public research layer.
My research sits at the intersection of ML, LLMs and computer vision, with a focus on inference optimization (KV-cache compression, quantization) and ensemble methods for large-scale forecasting. In my recent paper (IJCISIM 2026, DOI: 10.70917/ijcisim-2026-4333) I showed analytically and empirically that geometric log-space ensembling is the MAPE-optimal aggregator for log-normal targets, reaching MAPE ≈ 4.77% on an 18,000-store retail forecasting challenge. Open directions: efficient LLM inference, vision-language models, and transferring multiplicative-error ensembling to new domains
LLMComputer VisionOptimizationML