How it uses Jev
The app analyzes pasted WeChat-format text sentence by sentence with Jev, asking for emotion and intent classifications together, then rating the user’s replies and deriving a weighted set of relationship signals from recent text and relevant history.
What Jev decides
Is there enough text evidence to judge this relationship dimension?
Relationship task, selected message IDs, recent sender/text messages, relationship category and optional memory.
- sufficient
- limited
- insufficient
App workflow
- Paste and identify speakers
Paste copied chat text and select which nickname represents the user.
- Analyze the conversation
Process messages in batches, identifying emotion and intent and evaluating the user’s replies.
- Compute contextual indicators
Combine six weighted signals and relevant recent history into a displayed score and suggested next step.
- Continue with new messages
Append later chat excerpts, skip duplicates, and update the analysis while retaining prior local records.
Why it is interesting
It uses structured judgments for message-level analysis and keeps chat records in the current browser’s local database. The README cautions that the score is not a probability of affection and that analysis text is sent to TypeSafe.