Oleksandr K.

I help product teams and studios run image, video and vision AI on their own GPUs: diffusion, LoRA, video generation and computer vision pipelines in production, plus self-hosted LLMs and agents. 15 years in software. GPU-accelerated multimedia & vision Image and video generation, editing, restoration and analysis, built as automated pipelines rather than one-off outputs Computer vision and 3D: detection, tracking, segmentation, OCR and reconstruction for real-world video and imagery Model training and fine-tuning: LoRA, custom detectors and classifiers, tuned against measured targets High-performance inference: batching, quantization, GPU memory and throughput optimization, with cost per item measured, not guessed Self-hosted AI: open models on your own or rented GPUs (RunPod, Vast, GCP, AWS) in place of per-credit APIs AI agents & LLM systems Private LLMs on your own GPUs: vLLM serving, quantization, fine-tuning, OpenAI-compatible APIs Multi-agent workflows (LangGraph, Claude Agent SDK, MCP) and RAG with citations Evaluation harnesses, tracing and guardrails, so agents stay reliable Background Production generative-AI backends with GPU serving and concurrent inference APIs for several SaaS products First author of dnaSORA (bioRxiv 2025), a diffusion transformer for genomic data Agentic AI framework for a Mayo Clinic startup program participant MLOps: Docker, CI/CD, Terraform on GCP and AWS; Python, Rust, CUDA-level performance work How I work: I build a working sample on your own kind of data before we agree on scope, then deliver a documented, reproducible system your team can run without me. Not sure where to start? Book my 30-min consultation: you'll leave with an architecture, plan and estimate.