Marcel Osmond
Research, product, and legal systems
BLOG / FIELD NOTES
Marcel and Thomas log the build.
Essays, research notes, tiny product decisions, and behind the scenes updates from Marcel and Thomas as NYRVEN takes shape.
ENTRY 01
Why reliable AI should feel boring in production
The best production systems do not feel dramatic. They answer consistently, fail clearly, and leave enough evidence for a human team to understand what happened. For us, reliability means fewer surprises: grounded retrieval, visible citations, measured confidence, and evaluation loops that keep improving after launch.
ENTRY 02
Designing Aurora as a trust-first study companion
Aurora is built for students and institutions that need help they can inspect. The product should feel calm, useful, and honest: it retrieves the right material, explains where an answer comes from, and makes uncertainty visible instead of hiding it behind confident language.
ENTRY 03
What we learned building with evaluations from day one
Evaluations are not a final polish step. They shape the product from the first prototype. When every feature has to pass retrieval, citation, and answer-quality checks, collaboration gets easier: researchers, engineers, and partners can discuss evidence instead of vibes.
The next chapter of reliable AI is yours.
NYRVEN © 2026