LandPPT Jev Layout Selection
A presentation builder uses Jev to score slide-to-layout fit while deterministic assignment enforces layout constraints.
Added to Jevfast
How it uses Jev
When semantic selection is enabled and configured, LandPPT sends slide data and candidate component data to Jev. Each slide-component pair gets a Noul question about semantic fit; the returned Noul values score candidates for deterministic assignment, with LLM or rules filling incomplete results.
What Jev decides
Slide-to-layout fit (illustrative request shape)Example answers · not a recorded Jev response · Source ↗
Question 1 · q0
YOUR APP
INSTRUCTION
Does the selected slide semantically fit the selected component? Judge intent, relation, and information hierarchy; parallel arguments are not a sequence, a section divider is not a content page, and capacity alone does not imply fit.
STATE
A batch contains serialized slide records under state.slides and candidate component records under state.components. Pairing is representative: slide and component IDs are drawn from the batch, and the request asks one question for each pair.
JEV · NOUL
YesNo
App workflow
- Build legal candidates
LandPPT filters components for each page using content-capacity and image constraints before semantic scoring.
- Score slide-component pairs
With Jev selected and configured, it builds bounded batches of Noul questions and validates that each response contains a finite value from 0 to 1 for every pair.
- Complete and assign layouts
Missing Jev scores are filled using the configured LLM or deterministic rules; assignment then considers fit, image budget, and recent component history.
Why it is interesting
The code separates semantic fit judgments from layout assignment constraints: Jev evaluates each legal candidate pair, while beam search handles image budgets, repetition penalties, and layout continuity.