~/talk

From 6 months to 2 days: the LLM revolution for document processing

TalkinFrenchat Agile en Seine & IA 2026 · 2026-09-22..09-23· [large-language-models, machine-learning]
slides:none on file

A multimodal LLM replaces a months-long ML pipeline for document classification and data extraction (KYC: ID cards, passports, bank statements) by using prompts, data schemas, and business rules to convert raw PDFs to structured JSON in days without training data. The talk demonstrates a live end-to-end demo, compares the old CNN+OCR approach to the new prompt/schema/rule paradigm, shows production metrics (precision >99% after rules, cost, latency) and explains production considerations like hallucinations, degraded documents, observability, drift detection, DLQ, sovereignty and GDPR.

AI-written summary · about

~/search
type to search 6,518 conferences, 143,918 speakers and 192,133 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
← →
on a talk: 5 seconds back / forward
0–9
on a talk: jump to 0%–90% of the recording
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.