Oleksii Spesivtsev

Built scalable AI backend systems and multimodal pipelines to accelerate personalized video generation, reduce rendering latency, and streamline creative workflows. Delivered LLM-driven systems to extract plot elements, emotional nuances, and character dynamics from multimedia scripts for improved content personalization. Developed advanced scene segmentation and voice intent recognition models to enable dynamic visual storytelling. Led end-to-end development of a multimodal AI pipeline integrating LLaMA 2, Stable Diffusion, Whisper, and TTS technologies for multimedia automation. Built a real-time low-latency backend (Django + Channels, <200ms) supporting collaborative video editing with React/Next.js frontend integration. Designed and implemented a FAISS-powered asset retrieval engine improving media asset search relevance and editorial efficiency.

Led the full-stack development of a real-time AI-powered call center platform using Next.js (frontend), Django (admin & auth), and FastAPI (AI services). Architected backend services and database schema in FastAPI with PostgreSQL for scalable performance across thousands of daily interactions. Designed and implemented intuitive UIs in Next.js with TailwindCSS, integrating real-time updates via WebSockets. Engineered real-time transcription with FasterWhisper, semantic search pipelines with Pinecone, and LLM-driven intelligent query resolution using Google's generative models. Developed and deployed RAG pipelines for accurate, context-aware customer support responses. Built a smart escalation system for unresolved queries to human agents that preserved context. Led AWS cloud deployment with EC2, S3, Lambda, and CI/CD pipelines for high availability and cost-efficient scaling. Additionally contributed to Wrepit.co in developing risk detection and semantic search pipelines using RAG archi

Built scalable AI backend systems and multimodal pipelines to accelerate personalized video generation and streamline creative workflows. Architected LLM-driven systems to extract plot and emotional nuances from multimedia scripts, enhancing content personalization and automation. Developed scene segmentation and voice intent recognition models to accelerate AI video generation. Led end-to-end development of a multimodal AI pipeline integrating LLaMA 2, Stable Diffusion, Whisper, and TTS technologies. Developed a real-time low-latency backend supporting collaborative video editing with React/Next.js frontend integration. Designed a FAISS-powered asset retrieval engine, markedly improving media asset search relevance and editorial efficiency.