Roman Borysenok

Accomplished Software Engineer with extensive production experience in building scalable distributed systems, now actively expanding into AI/ML engineering. Proven track record in transforming business-critical systems and leading cross-functional teams. Recently developed sophisticated AI projects utilising LangChain, vector databases, and LLMs, combining established software engineering practices with modern AI technologies.
My strengths include distributed systems architecture, cloud infrastructure management, and DevOps practices. Passionate about continuous learning and applying emerging AI technologies to solve complex engineering challenges while focusing on production-grade reliability and scalability.

Portfolio

• AI-Powered Recruitment Platform

o Developed a multi-model AI system for candidate-job matching using LangChain, OpenAI, and vector embeddings.
o Implemented production-grade semantic search and advanced LLM chains for handling multi-step recruitment workflows.
o Achieved advanced feature engineering with custom prompt templates and chain orchestration.
o Tech stack: Python, LangChain, FAISS/Pinecone, FastAPI, Redis, PostgreSQL, AWS S3.

• AI-Powered Options Trading System

o Architected and implemented a multi-model ML system for options trading criteria evaluation using LSTM, Dense Neural Networks, and XGBoost.
o Developed innovative Positive-Unlabeled Learning approach for handling imbalanced financial data.
o Implemented advanced feature engineering with polynomial interactions and time-series analysis.
o Achieved significant improvement in trading decision accuracy (>90% precision) through hybrid model architecture.
o Tech stack: Python, TensorFlow, Keras, XGBoost, Pandas, NumPy, AWS S3.

• AI-Powered Image Narrative Generator

o Developed a full-stack application that transforms images into creative narratives using computer vision and LLMs.
o Implemented image captioning using BLIP model (Hugging Face) and narrative generation via GPT-4o.
o Integrated text-to-speech functionality using gTTS for accessibility.