Beyond the Model: Designing Reliable AI Agents

Why some AI agents actually work and most do not Same brain. Different harnesses. Wildly different results. If you've ever used an AI agent that confidently gave the wrong answer, this talk is about why that happens. Everyone discusses which model is "the best," but in practice, the model is rarely what makes an AI product succeed or fail. What’s often more important is everything around the model, the systems that determine whether it is reliable or embarrassing. Yaqoob and Veeresh, who have years of experience in data science and AI, will outline this in simple language and demonstrate live: the same AI model, two different setups, and wildly different results. Schedule 18:00 Doors open. Grab a drink or a snack and get to know the space. 18:30 Talk and live demo begins 19:30 Q&A 20:00 to 21:00 Drinks and conversation About the Speakers Veeresh Dammur is a Senior Data Scientist at Lovehoney Group in Berlin, with a background in natural language processing and machine learning from the University of Passau. He builds and manages the AI systems behind the company's ecommerce customer experience, including a production customer service robot that hands off conversations to human support agents and a product recommendation engine. Last year, much of his work has shifted from the models themselves to the machines around them, including agent harnesses, custom skills, evaluations, and automated end-to-end tests. https://www.linkedin.com/in/veeresh-prakash-dammur-1273157b/ Yaqoob David is a Senior AI Engineer at PlentyONE, where he builds AI-driven features for one of Europe's leading ecommerce platforms. He has an academic background from the Berliner Hochschule für Technik, also known as BHT. Before working at PlentyONE, he worked at Lovehoney Group, where he built and launched four production AI systems in seven markets, including a customer service assistant, a recommendation engine, and an automated editorial pipeline. His rule of thumb is simple: if it is not reliable and observable, it is not finished. https://www.linkedin.com/in/yaqoobdavid/ Between them, Yaqoob and Veeresh have built AI systems that have gone into production over the past few years, including customer service bots, recommendation engines, and editorial pipelines across multiple companies and markets. They keep running into the same problem: everyone is obsessed with which model to use, but the model is rarely what determines whether an AI agent works or falls apart in production. What matters is everything around the model: how it remembers information, what it is allowed to do, and what catches errors when something goes wrong. They call this the harness, and much of the difference between an impressive AI demo and a reliable AI product lies in how that harness is designed. In this talk, they will share what they have learned from building and launching these systems in real-world environments, including a few costly lessons. They will then demonstrate the concept live: the same underlying model, configured in two different ways, produces very different results. Same brain. Different harness. Ojin AixHaus: Ojin AixHaus: Berlin's center for human-centered AI innovation AI development is moving at a remarkable pace, but technical tools are only as powerful as the people who use them. Ojin AixHaus is a dedicated Berlin center designed to foster human connection, strategic execution, and the exploration of practical AI applications. It is a curated environment where AI enthusiasts, engineers, founders, and creative builders can go beyond the hype cycle and develop practical, real-world expertise. By offering free residencies for startups to organize sprints, events, workshops, and hackathons, Ojin AixHaus helps members and founders turn their ideas into action within a thriving community of experts. Come for the talk and stay for the live demo and the conversation afterward. The event brings together a relaxed and diverse audience of engineers, product professionals, founders, and people simply curious about AI, with plenty of time to meet others over drinks. 📅 Wednesday, July 29, 18:00 to 21:00 📍 Ojin AixHaus, Chaussee 36, 10115 Berlin 🎯 Open to everyone, technical or non-technical. Curiosity about AI is all that is needed. Discover more: Ojin AixHaus Community Follow us on social media: Ojin AixHaus Community Accessibility Please note that the venue is only accessible via stairs. There are currently no ramps or lifts available. Anyone with questions or concerns regarding accessibility is encouraged to reach out to the event organizer or the Ojin team prior to the event. Photo and Video Notice Photos and videos will be taken during the event. By attending, participants acknowledge that they may be photographed or recorded. Anyone with concerns should speak to the organizer or a member of the Ojin team before or during the event.