Zoran Marić
AI Solutions Architect
Preko 15 godina u razvoju softvera — od Laravel i Python servisa do AI agenata i LLM proizvoda. Spajam klasično inženjerstvo sa AI-em da napravim proizvode koji su i tehnički čvrsti i stvarno korisni. Over 15 years building software — from Laravel and Python services to AI agents and LLM products. I combine classic engineering with AI to build products that are both technically sound and genuinely useful.
Spajam back-end inženjerstvo i AI. Volim jasne arhitekture, pouzdane sisteme i praktičnu primenu agenata i LLM-a u realnim proizvodima. I combine back-end engineering with AI. I like clear architectures, reliable systems and practical use of agents and LLMs in real products.
AI Solutions Architect i softverski inženjer — projektujem inteligentne proizvode koji rešavaju stvarne poslovne probleme. Više od petnaest godina gradim web aplikacije, API-je i enterprise softver kroz ceo životni ciklus: od ideje i arhitekture do produkcije i dugoročnog održavanja. I'm an AI Solutions Architect and software engineer who designs intelligent products that solve real business problems. For over fifteen years I've built web applications, APIs and enterprise software across the full lifecycle — from idea and architecture to production and long-term maintenance.
Back-end iskustvo mi je temelj za pouzdane sisteme koji se skaliraju; danas je fokus na primeni AI-a za praktična rešenja. AI mi nije dodatak zakačen na softver — to je novi način da se softver gradi. Rad mi se vrti oko AI agenata, velikih jezičkih modela, inteligentne automatizacije i moderne arhitekture. My back-end background is the foundation for reliable, scalable systems; my current focus is applying AI to build practical solutions. I don't treat AI as a feature bolted onto software — I treat it as a new way of building it. Today my work centers on AI agents, large language models, intelligent automation and modern architecture.
Volim da složene probleme rešavam jednostavnim, elegantnim sistemima — AI asistent, automatska obrada dokumenata, multi-agent tok rada ili skalabilna cloud aplikacija. Tehnologije (PHP, Laravel, Python, Flask, Django, SQL, Docker) su samo alat; pravi posao je razumeti problem i izabrati najjednostavniju arhitekturu koja može da raste. I like solving complex problems with simple, elegant systems — an AI assistant, automated document processing, a multi-agent workflow or a scalable cloud app. Technologies (PHP, Laravel, Python, Flask, Django, SQL, Docker) are just tools; the real work is understanding the problem and choosing the simplest architecture that can evolve.
Uz razvoj proizvoda istražujem AI agente, enterprise AI arhitekturu i saradnju čoveka i AI-a, i pišem o spajanju klasičnog softverskog inženjerstva sa novom generacijom AI-native aplikacija. Alongside product work I research AI agents, enterprise AI architecture and human–AI collaboration, and I write about bridging traditional software engineering with the next generation of AI-native applications.
AI, agenti & LLMAI, agents & LLMs
Back-end & jeziciBack-end & languages
Baze, API & arhitekturaDatabases, APIs & architecture
DevOps & okruženjeDevOps & environment
Rečnik AI izraza — LLM, token, prompt, agentA glossary of AI terms — LLM, token, prompt, agent
Pedeset pojmova koji se stalno pominju, a retko objasne. Svaki ima jednu rečenicu odgovora, par redova konteksta i jednu sliku koja ostaje u glavi, a većina i link na lekciju u kojoj je razrađen. Fifty terms that come up constantly and get explained rarely. Each with a single-sentence answer, a few lines of context and one concrete image that makes it stick; most also link to the lesson where it's worked through.
Model ispod agenta — kako radi jezički modelThe model under the agent — how a language model works
Počni odavde ako nikad nisi direktno zvao LLM API. U 12 lekcija: jedan poziv, tokeni, prozor konteksta, tool use, structured outputs, izbor modela i cena. Temelj za celu agentsku seriju. Start here if you've never called an LLM API directly. In 12 lessons: a single call, tokens, the context window, tool use, structured outputs, model choice and cost. The foundation for the whole agent series.
Prompt i kontekst — šta staviš u modelPrompt & context — what you put into the model
Ne treba ti jači model — treba ti jasnije postavljen zadatak. U 12 lekcija: anatomija prompta, primeri, granica između uputstva i podataka, kontekst kao budžet, dovlačenje znanja i merenje promptova evalima. You don't need a stronger model — you need a clearer task. In 12 lessons: prompt anatomy, examples, the boundary between instructions and data, context as a budget, retrieval, and measuring prompts with evals.
Agentska petlja — kako agent zapravo radiThe agent loop — how an agent actually works
Interaktivni vodič u 12 lekcija: kako AI agenti rade kroz petlje, od mentalnog modela do radnog agenta sa Claude API-jem. Za programere nove u agentima. A 12-lesson interactive guide: how AI agents work through loops, from the mental model to a working agent with the Claude API. For developers new to agents.
Agent bez nadzora — memorija, ograde, evaliThe unattended agent — memory, guardrails, evals
Nastavak u 12 lekcija: memorija, kontekst, ograde, čovek u petlji, observability, evali i orkestracija — sve što razdvaja demo koji radi jednom od sistema koji radi hiljadu puta bez tebe. A 12-lesson sequel: memory, context, guardrails, human-in-the-loop, observability, evals and orchestration — everything that separates a demo that works once from a system that runs a thousand times without you.
Agent kao proizvod — MCP, servis, korisniciThe agent as a product — MCP, a service, users
Završni deo u 12 lekcija: pravi alati umesto igračke, MCP (alat kao servis koji koristi i Claude Code), agent iza HTTP servisa sa stanjem po korisniku, orkestracija na skali, i cost/latencija koji odlučuju da li uopšte ide u primenu. The final 12-lesson part: real tools instead of a toy, MCP (a tool as a service Claude Code uses too), the agent behind an HTTP service with per-user state, orchestration at scale, and the cost/latency that decide whether it ships at all.
Otvoren sam za saradnje na AI proizvodima i agentskim sistemima, ali i na klasičnim web, API i produktnim projektima — od prve ideje i arhitekture do produkcije i održavanja. Ako imaš problem koji AI može stvarno da reši, ili ti samo treba neko da čvrsto sklopi sistem, tu sam.I'm open to collaborations on AI products and agent systems, but also on classic web, API and product projects — from the first idea and architecture to production and maintenance. If you have a problem AI can genuinely solve, or you just need someone to build a system that holds up, I'm here.
Zanimaju me i kratke konsultacije i dugoročne saradnje. Najlakše me je dobiti mejlom ili preko LinkedIn-a — javljam se brzo.I'm open to both short consultations and long-term work. Easiest to reach me by email or on LinkedIn — I reply quickly.