Личные данные скрыты

Volodymyr Shabat
Remote, Ukraine | [открыть контакты](см. выше в блоке «контактная информация»)
LinkedIn | GitHub

Professional Summary
AI Systems & DevOps Engineer bridging the gap between production infrastructure and advanced AI deployment. Proven
track record of architecting cost-efficient Text-to-SQL pipelines (90% cost reduction) and optimizing inference on
constrained hardware (AMD/Vulkan). Currently managing network observability for 50+ ISP nodes while conducting
systems-level research in heterogeneous compute. Specialist in Linux kernel tuning, driver debugging, and agentic workflows.

Technical Skills
AI & Inference: Llama 3.x, llama.cpp (Vulkan/ROCm), Quantization (GGUF), RAG, Text-to-SQL
DevOps & Cloud: GitLab CI/CD, Docker, n8n, Nginx, Proxmox, VMware ESXi
Systems & Networking: Linux Kernel Tuning, MESA/RADV, Zabbix, Grafana, SNMP, Cisco IOS, MikroTik
Languages: Python (FastAPI/Django/Pandas), SQL (PostgreSQL), Bash, C# (.NET)

Professional Experience
AI Research Engineer (Independent R&D) Remote
Self-Employed / Research Sep 2024 – Present
– Engineered a production-grade inference environment for Llama 3 (8B) on unsupported AMD BC-250 mining
hardware by compiling custom MESA/RADV drivers and tuning Linux kernel parameters.
– Achieved stable inference at 32 tokens/sec, proving viability of low-cost hardware for edge AI.
– Architected MAC-SQL, a multi-agent system translating natural language to SQL. Replaced GPT-4 with
fine-tuned Llama 3.1 70B, reducing inference costs by 90% while maintaining 54% accuracy on BIRD
benchmarks.
– Built and released BIRD-UKR, a custom validation dataset, and implemented a reproducible evaluation pipeline
using Python and Docker.
DevOps & Infrastructure Engineer Kyiv Region, Ukraine
Baryshivka.NET ISP Aug 2024 – Present
– Engineered a centralized monitoring stack using Zabbix and Grafana for 50+ network nodes (Cisco/MikroTik).
Developed custom SNMP templates to visualize latency, reducing incident response time by 40%.
– Designed a low-latency Speech-to-Speech support agent by integrating Asterisk PBX with Ultravox and
ElevenLabs via WebSockets, solving real-time buffering challenges.
– Migrated internal tooling to a containerized Docker architecture and built GitLab CI/CD pipelines.
– Deployed n8n workflows to orchestrate data synchronization between billing systems and CRM.

Education & Training
Taras Shevchenko National University of Kyiv Kyiv, Ukraine
Bachelor of Science in Computer Engineering Jun 2025
– Thesis: ”Multi-Agent Systems for Database Interaction” (Implemented the MAC-SQL system).

Key Projects
Enterprise Infrastructure Automation | C#, .NET, Active Directory
– Developed C# automation tools to manage user provisioning and directory services via LDAP, reducing manual
onboarding time for system administrators.
– Deployed a resilient Windows Server environment (AD, DNS, IIS) within a virtualized infrastructure.