EBCONT at Weblica 2026: Smarter AI architecture. Lower costs.

The challenge is real: token consumption is growing exponentially, while chip production and GPU capacity can’t keep up. HBM memory is already reserved by major players through 2028, and memory costs are expected to rise by more than 50% by 2027. The typical response—cheaper models, shorter prompts—is a workaround, not a fix.

Omar presents a different approach: a multi-layered, fully local RAG architecture designed to reduce token load before the LLM even comes into play. Each component in the pipeline has a single purpose—to ensure the model only processes what it truly needs. He’ll demonstrate this using a production system from the energy sector.

The talk also addresses the other side of the same challenge: data sovereignty. For organizations operating under GDPR and the EU AI Act, sending internal data to external APIs isn’t an option. A well-designed local RAG setup solves both challenges at once.

Omar Drljevic and Bruno Šimić will be on site and look forward to connecting with the Croatian tech community.

Omar Drljevic, AI & Data Consultant at EBCONT, will take the stage at 12:00 with his talk: “Više AI-a, manje tokena: Optimiranjem arhitekture do manjih troškova” — in English: “More AI, fewer tokens: Reducing costs through optimized architecture.”

Want to learn more about our work in AI & Data or connect on site? Reach out directly to Bruno: bruno.simic@ebcont.com

All conference details: weblica.hr