Chief Artificial Intelligence Architect (Solo Builder)
26 years in enterprise infrastructure and AI systems. Solo builder developing AI platform from ground up through AI-assisted development. Work spans Unix/Linux administration, cloud-native platforms, and AI application development. First production rollout in progress (Q1 2026). Recent focus: agent orchestration, RAG systems, multi-provider LLM gateway, real-time voice AI with telephony integration.
Built entirely through AI-assisted development with infrastructure-grade discipline. Evidence-based integration (inspect library internals, cite actual behavior), full type coverage (basedpyright strict), automated quality gates. Multi-agent orchestration patterns with state management, system prompts, handoff protocols.
Built systems using pgvector (PostgreSQL), Qdrant, Milvus. Multi-space vector architectures, embedding strategies, query refinement, conversation history management. E-commerce chatbot with 100,000+ products going live Q1 2026.
Built voice agents using LiveKit VoicePipelineAgent with Asterisk/SIP telephony integration. STT/TTS pipelines (Deepgram, ElevenLabs, Cartesia), voice activity detection, call transfer, multi-language support. Proof-of-concept development.
Kubernetes production clusters with NFV capabilities for telecommunications. OpenShift (3.11, 4.x) at enterprise scale. 26 years with Linux (RHEL, Ubuntu, SUSE), 15 years with UNIX. Security-cleared environments (NATO, eu-LISA European institutions).
AI agent orchestration platform with multi-tenant architecture. Event-sourced architecture with deterministic replay and audit. Multi-agent workflows with manager→specialist delegation, handoffs, bounded parallelism. Governance layer: approval workflows for high-risk operations, policy-based tool execution, RBAC. Platform components: Operator Console, Customer Portal, Python SDK/CLI, embeddable chat. Provider abstraction via whitelabeled engines—zero OpenAI/Anthropic/Google exposure at API edge. Token-based billing, cost controls, quality gates with eval sets and red-teaming. Under active development.
Production deployment for Balkans health and wellness operation. MCP server with multi-space vector architecture (health conditions, ingredients, beauty concerns, guides) across 100,000+ products. pgvector HNSW indexes, real-time streaming chat, Magento OAuth integration, admin dashboard, conversation tracking, Serbian character preservation. First production rollout Q1 2026.
Real-time voice agent for telemonitoring support. LiveKit VoicePipelineAgent with Asterisk/SIP telephony integration. STT/TTS pipelines (Deepgram, ElevenLabs, Cartesia), voice activity detection, call transfer, multi-language support. Proof-of-concept and platform integration development.
Unified SDK across OpenAI, Anthropic, Google, xAI, Ollama. Whitelabeled architecture with zero provider exposure in public APIs. Streaming-first design with SSE, engine-based routing, customer API key management. Production infrastructure for multi-provider orchestration.
Built entirely through AI-assisted development with infrastructure-grade discipline. Methodology emphasizes evidence-based integration (inspect library internals, cite actual behavior), full type coverage (basedpyright strict, zero type ignores), automated quality gates (Ruff, complexity analysis, test coverage). Core principles: minimalism, explicit structure, orthogonal design. Multi-agent orchestration patterns with manager→specialist delegation, state management, system prompts, handoff protocols. FastAPI vertical slice patterns. Solo execution.
Value-Shore | Belgrade
IBM (eu-LISA) | Strasbourg
Remote
ITDM Srl | Rome
Multiple Clients | Italy
Enterprise Clients | Europe
Harpa Italia | Italy