Software Architect · AI Engineer · Zagreb, Croatia
I build AI systems that
survive production.
I'm Tihomir Tomašević — a software architect and engineering leader with 17+ years delivering enterprise-scale systems across telecom, government, and SaaS. For the past four years I've been designing and leading development of an internal agentic AI platform with deep LLM integration.
T2 Software is my consultancy. I take on selected contract and advisory work for teams putting LLMs into real systems — the part where prototypes usually break.
17+ years building systems at
What I do
Three things, all aimed at the same problem: getting AI out of the notebook and into a system that a business can actually depend on.
Agentic system design
Multi-agent architectures, tool-calling, orchestration and memory design. The hard part is rarely the model — it's the control flow, the failure modes, and knowing when an agent is the wrong answer.
LLM evaluation & observability
Eval suites, tracing, prompt versioning, cost and latency tracking. I spent years building business-activity monitoring for telecom before LLMs existed; the discipline transfers directly.
Backend & cloud architecture
Microservices, event streaming with Kafka, PostgreSQL and vector stores, AWS/GCP, CI/CD. The unglamorous foundation that decides whether your AI feature holds up under load.
Why an enterprise background matters for AI
Most of the hard problems in production LLM systems are not new problems. They're distributed-systems problems wearing a new hat: retries, idempotency, partial failure, observability, cost control, schema drift.
- I've run the full lifecycle. Architecture, delivery, and a 25-engineer team — not just prototypes.
- I've shipped under real constraints. National-scale public sector systems, telecom monitoring, regulated environments.
- I know when not to use AI. Sometimes the honest answer is a queue and a state machine.
Working on something in this space?
Whether you need an architecture review, a second opinion on an agent design, or someone to build the thing — I'm happy to talk. No pitch deck.