At AquaBioSafe Laboratory, I built an open-source computer-vision cell counter for an international harmful-algal-bloom monitoring project. It reduced sample counting from hours to minutes with almost no manual work, reaching 5.3% mean count error on 12 annotated test images.
I also developed a local RAG assistant for protist identification. It searches about 110 papers, cites supporting quotes for candidate taxa, and returns “insufficient evidence” when retrieval is weak.
Across three Russian Science Foundation projects and a World-Class Research Center program, I’ve worked on metabarcoding, metagenomics, genome and transcriptome assembly, differential expression, and phylogenetics. I also administer the lab’s Linux GPU server, which holds more than 50 TB of data.
At Polyus R&D Center, I built Random Forest and logistic regression models to forecast gold recovery, helping move the team from descriptive reporting to prediction. My work cut weekly sample-analysis time by 15% and monthly reporting by four hours.
