WhatsApp Catch-Up

The problem

Voice notes are easy to send and expensive to catch up on. Decisions, deadlines, and questions can be scattered across several recordings. A receiver has to listen, remember, and decide what needs a response.

The product bet

Make catch-up a recovery layer inside the conversation. The central interaction is a short loop: skim, verify the source, edit a reply, and send deliberately.

I prioritized source traceability and user control over autonomous replies. A summary should help the user understand the conversation before it asks them to act.

What works in the prototype

The main scenario contains seven voice notes totaling eighteen minutes. The build includes an inline summary, detail panel, global catch-up inbox, transcript search, playback controls, and reply helper.

Sending a reply updates the answered state. Topics, questions, decisions, confidence, and source timing are represented as typed data. A companion Product Sense Flow Map explains the feature logic and tradeoffs.

What is simulated

The AI outputs and conversation data are mocked. This is an interactive workflow prototype, not a live transcription or summarization service, and it does not connect to a WhatsApp account.

A production version would need a privacy architecture that preserves end-to-end encryption. On-device processing and explicit consent for any cloud inference would need evaluation.

How I would evaluate it

Compare time to identify the correct decisions and unanswered questions against listening to the original notes. Track missed asks, unsupported summary claims, source-verification behavior, and edits to suggested replies.

Include ambiguous commitments, conflicting dates, multiple speakers, and imperfect transcripts. Accuracy and recoverability should be checked before increasing automation. These are proposed evaluation criteria; no model-quality or user-study results are claimed.