Lead AI Engineer & Architect - Applied Agentic AI

A product worth your next few yearsSome of the most interesting AI engineering problems today are not in research labs. They’re in real products that need to work reliably, every day, for real users. Cempresso Next-Gen is one of them. We’re rebuilding a proven telecom diagnostics and troubleshooting platform around agentic AI.We’re looking for an experienced engineer to take over the new AI team behind it: to own the architecture, make the hard design decisions, grow a team, and see the work reach production with real subscribers on the other end.What is Cempresso?For 15 years, Bulb has been building Cempresso: the SW platform telecom operators use to diagnose and fix complex broadband, TV, voice and mobile service problems. It sits between the agent, subscriber or field technician and the operator’s entire technical estate, including access networks, CPE devices and OSS/BSS systems, and turns “my internet is slow” into a root cause and a fix.Next-Gen Cempresso is a ground-up evolution. The existing R&D team continues to own the core engine, integrations, admin tooling and agent console. You will own everything AI, and the two teams will build the product together.Not just another agentic orchestratorMost agentic frameworks point a model at loosely structured context and hope for the best. Cempresso’s bet is different: a heavily deterministic backbone, with a well-organized session data model, deterministically anchored key steps in the customer-service process, and structured context that agents read from and write to, with AI reasoning applied exactly where reasoning adds value.The target is to outperform custom and general-purpose AI orchestrators on what operators actually measure: accuracy and consistency of answers, elimination of hallucinations, and reliability across deep, complex integrations.The use case is diagnostics and troubleshooting of complex services. The ambition is a superior, universal, AI driven customer-service platform.The problems you’ll solveMulti-thread AI sessions. One troubleshooting session will run several parallel, context-specialized AI conversations, like opening separate chats for separate sub-problems, coordinated through shared, well-structured session context rather than each thread’s own memory.Context access at scale. Session context is large and heterogeneous: live data from network and infrastructure elements, OSS/BSS records, the history of previous sessions, external analytics, and the subscriber’s own account of the problem. Plain vector RAG won’t cut it. You’ll design the methods that give agents the right slice of context at the right moment, balancing accuracy, latency and token cost.Tool-rich agents. Agents will have a large toolset available via MCP, APIs and the platform’s integration framework. You’ll define how they discover, select and safely invoke those tools for advanced diagnostics and remediation.Knowledge at telco scale. Operators bring extensive knowledge bases; you’ll decide how that knowledge is structured, grounded and kept trustworthy.Self-service channels. You’ll shape how the AI performs across chatbot and voicebot channels: conversational quality, structured outputs and handover to humans.Self-learning and hybrid models. You’ll contribute to the platform’s ML component, learning from session outcomes to continuously improve diagnostics and decision-making.Your role· Architect and technical lead for all AI capabilities of Cempresso Next-Gen· Take over and lead the product AI team, with the opportunity to further grow and shape it· Hands-on: design, prototype, benchmark and ship, in close partnership with the core R&D team and telco domain experts· Beyond Cempresso: technical lead on other ongoing Bulb AI projects, with ownership of evolving the agentic modules and production chatbot installations Bulb has already deliveredWho we’re looking for· Deep, hands-on experience building LLM-based systems that reached production, including agent architectures, tool use, context and memory management, and evaluation· Strong engineering fundamentals; you’ve owned the architecture of complex systems and can defend design decisions with data· Real opinions about where different RAG approaches break, how to keep agents deterministic where it matters, and how to measure hallucination rather than just talk about it· Experience leading engineers, or a clear appetite to build a team· Comfort working alongside domain experts on problems you won’t fully understand on day one· Fluent English; Croatian or another regional language is a plusWhat you get· A product with real users and real stakes, with existing operator installations across the SEE region waiting for their next generation· Ownership of the AI architecture from day one, and a team you shape· A company that has invested in AI for years and has a portfolio of AI projects running in production· Compensation that reflects the seniority and scope of the role. This is a leadership position, and we treat it as one· A flexible work model, the equipment and tools you choose, and a benefits package built around long-term collaboration rather than perks on a posterHow to applySend us your CV and a short note about the most complex AI system you’ve shipped and what you’d do differently today. That note will tell us more than any list of technologies.Don’t let the requirements above stop you from applying. They describe the experience and skills we think are relevant to the role, not a checklist you need to match line by line.