Core Team
Pascal Bertrand
Founder, AI Startups and Collaborations
Research AssistantAbout
Pascal Bertrand is a physics student at ETH Zurich and an early builder working at the intersection of science, entrepreneurship, and agentic systems. Since joining the lab from its earliest days, he has helped shape its direction with a strong bias toward execution and real-world impact. His background combines analytical training in physics with hands-on experience in deep-tech engineering and venture investing, giving him an unusually broad perspective on how ambitious technologies move from concept to adoption. Before joining the lab, Pascal worked on rockets at ARIS and gained exposure to the startup ecosystem through Founderful, where he engaged closely with Swiss founders and emerging ventures. That mix of technical curiosity and founder instinct started early: as a teenager, he co-founded a company with his brother and generated meaningful early revenue. At the lab, he now focuses on building systems that do not just analyze information, but turn noisy, fast-moving real-world data into decisions and actions.
Publications
Research Areas
Project
Sentiments: Agentic Market Intelligence from Short-Form Video
Sentiments is an agentic system designed to extract actionable market intelligence from millions of short-form videos across social platforms. Rather than treating videos as unstructured noise, the system uses vision-language models, semantic retrieval, and reasoning components to identify themes, reactions, emerging preferences, and behavioral signals at scale. It is built as an end-to-end pipeline: from scraping and multimodal analysis to synthesizing recommendations and proposing concrete actions such as product adjustments, campaign refinements, or creator collaborations. The broader goal is to transform social media from a passive observation channel into an active decision engine for marketing and product teams. Scientifically, the project sits at the intersection of multimodal machine learning, large-scale information retrieval, and agentic reasoning. A central challenge is how to move from fragmented, highly contextual video signals toward robust higher-level conclusions without losing nuance. This requires systems that can interpret visuals, language, sentiment, and cultural context jointly, while also reasoning over time, geography, and audience segments. The project therefore contributes to broader questions around multimodal understanding in dynamic environments: how AI systems can detect weak signals early, distinguish transient noise from meaningful patterns, and generate explanations that remain useful to human decision-makers. Commercially, Sentiments addresses a growing need for organizations to understand markets in real time rather than through delayed surveys, static dashboards, or manually curated reports. Brands, consumer platforms, and agencies increasingly depend on rapid feedback loops, yet most existing tools stop at surface-level analytics. By combining detection, interpretation, and recommendation in one workflow, Sentiments points toward a far more operational layer of market intelligence. Its ability to monitor trends across target groups, regions, and communities makes it especially relevant for fast-moving sectors where product, brand, and distribution decisions need to be continuously updated from live public behavior.
Other team members
Students
Interested in collaborating?
We are always looking for talented students, researchers and industry partners.
