Mehedi Hassan is a London-based product engineer at Granola and co-founder of Sonder, a platform for forming friendships and relationships through expressive profiles and in-person events. He builds AI products that preserve individual judgment: meeting notes shaped by what users consider important, and social connections assisted by algorithms but grounded in human effort.
Before Granola, Hassan worked with Qualcomm, Circle, and Causaly. His independent projects include Tweeten, a customizable desktop Twitter client built around TweetDeck; Bulksplash, a command-line tool for downloading Unsplash images; and GitCleanup, which removes abandoned or empty GitHub repositories.
Building AI products people can actually trust
- Production feedback loops. Hassan treats a convincing AI prototype as the beginning of product engineering. Granola’s meeting assistant must handle ambiguous requests, expensive web searches, unexpected changes in model-provider behavior, and follow-up messages that reflect how particular users communicate. His AI Engineer talk shows how observing failures and testing changes repeatedly makes these systems dependable.
- Observability beyond engineering. Granola’s internal tracing tools expose tool calls, search paths, reasoning trails, and costs through interfaces usable by product, data, and customer-experience teams. This gives collaborators enough context to diagnose failures without navigating infrastructure logs.
- Browser previews for desktop applications. Hassan describes adapting Granola’s Electron renderer into a browser-accessible web shell by abstracting desktop integrations and using web-standard routing, sessions, and data access. Continuous integration can then generate shareable pull-request previews, while Cursor-assisted checks attach screenshots for faster verification.
- Personalization without impersonation. Granola combines meeting transcripts with users’ own notes, allowing outputs to reflect different professional priorities: sales opportunities, engineering blockers, or action items. At Sonder, which Hassan co-founded with Helen Sun, Lenard Pratt, and Hannah Kin, language models can suggest connections, but users create their own collage-like profiles. Hassan opposes AI-generated profiles because the effort of presenting oneself signals genuine interest in another person.
