Foresight’s AI Salon: Matin Mahmood | Hyperbolic Embeddings and Jev for Search
Hosted by Foresight Institute Calendar + Lydia La Roux + Bradley C Royes (Foresight Institute) + Matin Mahmood
Register with organizer
What to look forward to
Foresight’s AI Salon will feature hyper³labs founder Matin Mahmood, whose Hyper3-CLIP research earned an oral presentation at ECCV’s Beyond Euclidean workshop, exploring hyperbolic embeddings, model training, interpretability, and search applications. The evening will include a talk, Q&A, and breakout discussions on image–text retrieval, RAG, and open questions about the role of geometry in model performance.
Inside the event
Join Foresight’s AI Salon for an evening with Matin Mahmood, founder of hyper³labs, exploring hyperbolic embeddings, what they can tell us about hierarchy, and lessons from building and training models with them. Matin will introduce the geometric intuition behind hyperbolic representations: how a model’s embedding space can reflect relationships between broad categories and more specific concepts. Drawing on Hyper3-CLIP, hyper³labs’ hyperbolic image–text model research, he’ll share training lessons and image–text retrieval examples. The discussion will also address an open question in this work: how much of a model’s performance can be attributed to the geometry itself? Matin will also briefly explore interpretability in hyperbolic spaces: how relationships between concepts might help us understand what a model has learned. To close, Matin will explore how Jev can be used in search, with examples of selecting useful passages for RAG and reranking results against specific criteria. The evening combines a short talk with Q&A, breakout discussions and time to meet other researchers and builders. People working on model training, information retrieval, knowledge graphs and search are welcome, including anyone interested in research collaboration. About the speaker Matin Mahmood is the founder of hyper³labs, an AI research company based in Berlin. His work spans representation geometry, text embeddings and multimodal learning. The research behind Hyper3-CLIP received an oral presentation at ECCV’s Beyond Euclidean workshop. Previously an ML Scientist at GE Vernova, Matin holds a BSc (Hons) in Artificial Intelligence and Mathematics from the University of Edinburgh. When and where Wednesday, 14 October 2026 · 18:00–21:00 CEST CIC Berlin 18:00–19:00: Drinks and social time 19:00–19:20: Talk 19:20–19:40: Q&A 19:40–20:00: Breakout session 20:00–21:00: Social time Salon discussions are held under Chatham House Rule (don’t connect people to ideas when discussing them outside of this workshop). Foresight’s AI Nodes offer grant funding, local compute and community hubs for AI for Science and Safety projects in Berlin and the Bay Area.
Berlin, Germany
Venue listed by the organizer · Berlin, Germany. Confirm the final address before traveling.
Open in Google Maps ↗Discovered through Cerebral Valley. Details come from the event listing and the organizer page; registration is managed by the organizer.
Last checked: Oct 14, 2026. Confirm current availability before making plans.
Organizer page checked: Oct 14, 2026. Check registration for the latest details.