Oleksii Kaplenko

Senior AI/ML Engineer specialising in LLM-based and agentic systems, with commercial machine-learning experience since 2017. Today I mostly build LLM systems in production: multi-agent and tool-calling architectures, RAG, model routing across Claude, GPT and Gemini (LiteLLM), LLM evaluation (golden datasets, LLM-as-a-judge, drift and regression checks, CI eval gates), observability (Langfuse, OpenTelemetry) and guardrails.

At SaneAI I led the build of an AI strategic-planning SaaS for enterprises from scratch: architecture, backend (Python, FastAPI), frontend (React, TypeScript), database (Supabase/PostgreSQL) and CI/CD. I designed the LLM layer with cross-vendor fallback and versioned prompts, built agentic tool calling that lets users edit the plan through chat with guardrails and permission-scoped access, created the evaluation pipeline used as a release gate, and built scheduled autonomous agents with human-in-the-loop approval and AI cost caps. I also set up an AI harness for the SDLC (Claude Code subagents, skills, hooks, MCP servers) that turned the team AI-driven.

At TSGS, a proprietary trading firm, I architected a multi-agent portfolio-management system (LangGraph, CrewAI, AutoGen) where agents for risk, sentiment and signal generation reach a joint decision, fine-tuned open-weight LLMs with LoRA and served them with vLLM and SGLang, and ran production trading strategies with deep reinforcement learning (PPO).

At Digis I built an LLM customer-support assistant with LangChain and RAG over vector databases that cut manual support workload by 40%, a GPT marketing-content generator, and computer-vision features (image retouching, AR try-on, generative background replacement). At Thermo Fisher Scientific I led deep-learning projects for electron microscopy (super-resolution, segmentation, 3D reconstruction), which resulted in 4 granted US patents as first author.

Master's in AI and Machine Learning Systems (Brno University of Technology). Claude Certified Architect (Anthropic, 904/1000). English C1, Czech C1, Ukrainian native.

Looking for: fully remote, hands-on work on LLM agents, RAG, LLM evaluation, fine-tuning and inference; contractor/B2B preferred, employment also possible.