BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Scaling LLMs to serve the academic community - Ryan Daniels (Unive
 rsity of Cambridge)
DTSTART:20260210T141000Z
DTEND:20260210T142500Z
UID:TALK244270@talks.cam.ac.uk
DESCRIPTION:Modern AI research faces a crisis of dependency. As we increas
 ingly rely on commercial APIs and closed-source models\, the scientific co
 mmunity risks undermining the reproducibility\, interpretability\, and sus
 tainability of its work. This talk argues for a strategic pivot: moving fr
 om viewing AI as a tool we rent\, to viewing it as infrastructure we contr
 ol. We present a case study in building this "sovereign stack." Moving bey
 ond the traditional model where hard-won compute resources are necessarily
  siloed\, we demonstrate how we architected a server to operate as a share
 d utility for the academic community. However\, independence brings comple
 xity. We candidly explore the significant engineering friction encountered
  in moving from raw hardware to a production-ready service. We discuss the
  realities of hardening the attack surface\, managing API keys\, and the i
 ntricate balancing act of optimizing throughput\, latency\, and VRAM usage
  against massive context windows. By releasing our full software stack\, m
 onitoring configurations\, and Architecture Decision Records (ADRs)\, we a
 im to "outsource the tedium" of these discoveries. This talk serves as a b
 lueprint for how institutions can build secure\, scalable\, and open AI in
 frastructure that keeps science reproducible and data secure.
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR