Olga Poletkina

• Built a Revit-integrated ML assistant for semi-automated BIM quality checking. • Managed cloud deployments (Runpod, Yandex Cloud), optimizing GPU usage and costs. • Developed custom GNN + CatBoost models (85–95% F1; ~5 cm spatial precision). • Automated Revit family placement with GNNs, cutting design cycles from months to days. • Implemented RAG for technical documentation (FAISS + LLM reranking) for engineering teams.

• Built an automated Revit-to-cloud LLM pipeline enabling real-time predictions. • Developed ML for text-to-parameter mapping and room attribute classification. • Prototyped a corporate RAG system for documentation (sentence-transformers + FAISS). • Implemented custom reranking logic to improve retrieval accuracy.

• Delivered ML-enhanced BIM automation pipelines for data-driven project decisions. • Built Python automation frameworks (pyRevit/Revit API) for large-scale data extraction and visualization. • Created analytics tools (e.g., CO2 estimation, license utilization) and reduced Revit license consumption by 2x. • Trained a LoRA-adapted Stable Diffusion model to accelerate early-stage design ideation.

Led BIM coordination and architectural design for 14 years, driving the implementation of digital workflows across international project teams from Germany and Poland.

Conducting Master's research: Cascaded R-GCN architecture for automated fire safety compliance checking in BIM models, combining graph neural networks with regulatory knowledge graphs.