NewJevclip
Select useful transcript segments with Jev and assemble source-linked video highlights.
Added to Jevfast
NewHow it uses Jev
For each transcript segment, Jev receives the video title and that segment's text, without its timestamp, filename, or neighboring segments. One request asks a Choice about content kind, a Score about substance, and Noul questions about unsupported hype and whether the segment stands alone; up to five user focus points add Noul questions. Code combines the answers with a policy to decide what to keep and what to skip in the full version.
What Jev decides
Assess one transcript segment (illustrative request shape)Example answers · not a recorded Jev response · Source ↗
Question 1 · kind
YOUR APP
INSTRUCTION
Classify the segment as one of the rubric's eight content kinds.
STATE
The request state contains a video title and one segment's text. The repository does not prescribe a particular title or segment text, so this example describes the fields without inventing runtime values.
JEV · CHOICE
- insight
- evidence
- method
- explain
- story
- smalltalk
- promo
- filler
Question 2 · substance
YOUR APP
INSTRUCTION
Rate the segment's information density and cognitive value on the four-level rubric.
STATE
The request state contains a video title and one segment's text. The repository does not prescribe a particular title or segment text, so this example describes the fields without inventing runtime values.
JEV · SCORE
0
Question 3 · hype
YOUR APP
INSTRUCTION
Judge whether the segment makes exaggerated or unsupported claims based on this segment alone.
STATE
The request state contains a video title and one segment's text. The repository does not prescribe a particular title or segment text, so this example describes the fields without inventing runtime values.
JEV · NOUL
YesNo
Question 4 · standalone
YOUR APP
INSTRUCTION
Judge whether a viewer can understand the segment without the preceding context.
STATE
The request state contains a video title and one segment's text. The repository does not prescribe a particular title or segment text, so this example describes the fields without inventing runtime values.
JEV · NOUL
YesNo
App workflow
- Prepare transcript segments
Pair supported transcript formats with video files when available, segment the text, and retain source timing for reports and edits.
- Judge each segment
Build the rubric questions and send the title plus one segment per request. Reuse a cached answer when the same model, text, and question wording already have a result.
- Apply policy and edit
Code derives keep, skip, value, and reasons from the returned answers and configured thresholds. It orders retained segments for a highlights reel, builds the information-preserving full version, and writes a report and summary.
Why it is interesting
The project separates segment-level judgments from editing policy. It can change keep/drop thresholds without repeating model calls, preserves reasons for excluded segments, and keeps transcript wording tied to source timestamps for later editing and summary references.