BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//cfp.devcon.org//devcon8//talk//ETGLCN
BEGIN:VTIMEZONE
TZID:IST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1
TZNAME:IST
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-devcon8-ETGLCN@cfp.devcon.org
DTSTART;TZID=IST:20261105T143000
DTEND;TZID=IST:20261105T144500
DESCRIPTION:Visdom is FOSSASIA's open source dashboard for watching machine
  learning experiments live\, with over 10\,000 GitHub stars. Over five mon
 ths we rebuilt a large part of it in five stages\, with 92 pull requests m
 erged upstream\n\n1. Storage abstraction layer: a DataStore interface so e
 very environment save\, load\, delete and undo goes through one backend in
 stead of raw JSON files.\n\n2. Search layer: an experiment metadata store\
 , an injection-safe query parser\, search and field-by-field comparison.\n
 \n3. Hyper-parameter layer: a dashboard with a run table\, scatter matrix\
 , parallel coordinates\, a compare view and live updates.\n\n4. Pytest: a 
 suite of nearly 1500+ tests that uncovered real crashes and security bugs.
 \n\n5. asyncio: disk I/O moved off the event loop and a new AsyncVisdom cl
 ient\, tackling an issue open since 2020 and reaching 2.1x the old client'
 s throughput.\n\nI'll show how each stage made the next one possible\, the
  bugs we found along the way\, and why small\, reviewable PRs that never b
 reak the existing API are what got a mature project to accept this much ch
 ange.
DTSTAMP:20261101T051110Z
LOCATION:Community Led Stage
SUMMARY:Rebuilding Visdom AI Visualizations in Five Layers - Manik Khajuria
URL:https://cfp.devcon.org/devcon8/talk/ETGLCN/
END:VEVENT
END:VCALENDAR