Nikita Gaidamachenko

● Developed a B2B Ideal Customer Profile using K-NN and SASMOTE with MLRegressor for Autoencoding on a minority set with 198 features. Achieved accuracy of over 80%. Scraped and engineered a dataset based on TARION, permits, and other sources making over 500,000 API calls. Expanded API functionality for TARION by 3 times. ● Created a 6-step plan for senior leadership to implement an Agentic AI for Requests for Proposals (RFP), assess suitability and prepare proposal drafts. Led the team to develop Proof of Concept (POC) for B2B & B2C models.

● Reduced a CPA firm's data entry time by 68% by fine-tuning an LLM using HuggingFace and PyTorch, that extracts data from bank statements and invoices, formatting the output for immediate copy-paste into QuickBooks. ● Built a mortgage pre-approval likelihood model for a brokerage using XGBoost with distributed training on Google Cloud, incorporating TensorFlow-based data validation and deploying the model via Vertex AI for scalable inference, improving lead qualification accuracy by 32% and reducing manual reviews. ● Reduced HR workload by 52% by building a RAG-based policy Q&A system using Watson AI, Mistral, LlamaIndex, PyPDFLoader, and ChromaDB to automate employee policy inquiries.