BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.devconf.info//devconf-us-2026//speaker//FVUERM
BEGIN:VTIMEZONE
TZID:EST
BEGIN:STANDARD
DTSTART:20001029T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10;UNTIL=20061029T070000Z
TZNAME:EST
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
END:STANDARD
BEGIN:STANDARD
DTSTART:20071104T030000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11
TZNAME:EST
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000402T030000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=4;UNTIL=20060402T080000Z
TZNAME:EDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
END:DAYLIGHT
BEGIN:DAYLIGHT
DTSTART:20070311T030000
RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3
TZNAME:EDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-devconf-us-2026-KYRBPB@pretalx.devconf.info
DTSTART;TZID=EST:20260924T142000
DTEND;TZID=EST:20260924T145500
DESCRIPTION:In March 2026\, the LiteLLM package was compromised on PyPI thr
 ough a hijacked CI/CD pipeline\, silently exfiltrating credentials and sec
 rets. Months earlier\, fake DeepSeek packages targeted ML engineers with i
 nfostealers. Standard Python tooling resolves and builds packages individu
 ally with no unified view of the full dependency tree\, making it hard to 
 audit what was built\, from where\, and in what order.\n\nFromager solves 
 this by capturing the entire dependency tree as data. A bootstrap step rec
 ursively resolves all dependencies and outputs the full graph and build or
 der as plain JSON artifacts you can review\, diff\, and version-control be
 fore any compilation happens. From that plan\, every package is rebuilt fr
 om source in the correct order\, optionally with network isolation to cons
 train what build scripts can reach. The result is a fully visible\, reprod
 ucible supply chain.\n\nFromager builds packages as collections\, the full
  dependency tree is compiled together\, so packages linked against each ot
 her stay ABI-compatible. This matters for stacks like PyTorch and vLLM whe
 re mismatches between CUDA or ROCm binaries cause silent failures.\n\nA pl
 ugin and override system lets you customize each build stage (download\, u
 npack\, patch\, build) per package. One upstream recipe can produce builds
  for different hardware targets while keeping the rest of the configuratio
 n shared.\n\nWe'll cover what we've learned building 1000+ packages for pr
 oduction AI workloads: dependency patterns specific to AI stacks\, where s
 ource builds break at scale\, and why building as collections is necessary
  for ABI correctness across accelerators.\n\nThis talk is for engineers wh
 o build\, deploy\, or audit Python environments for production AI workload
 s\, especially those dealing with hardware-specific builds\, reproducibili
 ty requirements\, or supply chain compliance. Attendees will leave with an
  understanding of how to evaluate their own Python supply chains and the t
 ools to start building from source themselves
DTSTAMP:20260727T165155Z
LOCATION:106 (Capacity 45)
SUMMARY:From Source to Production: Securing Python AI Stacks Against Supply
  Chain Attacks - Rohan Devasthale\, Lalatendu Mohanty
URL:https://pretalx.devconf.info/devconf-us-2026/talk/KYRBPB/
END:VEVENT
END:VCALENDAR