Daniil Krechko

Remote · client: medical imaging company • Develop a modular AI pipeline that ingests DICOM studies from PACS, runs anatomy-specific models and returns DICOM SR/SEG reports. • Lead the brain MRI module, defining KPIs and report fields for lesion segmentation and volumetry. • Audited clinical training data, uncovering unusable annotations and missing labels; redesigned the data strategy. • Assessed commercial licensing of reference models (MONAI, BraTS) and defined a compliant model and dataset strategy.

Remote • Led end-to-end development of training and inference pipelines for Stable Diffusion–based systems used by real users. • Monitored output quality via CLIP-embedding similarity, maintaining 90–95% style similarity thresholds. • Implemented automated dataset restructuring pipelines, achieving 100% compatibility across diverse user-provided datasets. • Built user-facing GenAI workflows translating customer requirements into accessible AI solutions.

Remote • Automated data ingestion and preprocessing pipelines, replacing manual SQL-based workflows. • Applied ML models to analyze 50k+ textual reviews, generating structured insights and feature-based reports. • Evaluated and tuned multiple ML approaches with focus on accuracy and robustness.

Bachelor thesis: Multimodal Recommender Systems with a Focus on Cold-Start Problems. GPA 2.0 (1 = excellent). Projects: BraTS-Glioma segmentation on multi-modal MRI – 3D U-Net (MONAI), best result among course teams (Mean Dice 0.80). Multimodal music retrieval – Top-POP, SVD, ItemKNN on implicit feedback; nDCG@10 improved from 0.12 to 0.17.