Inner Voice: our AI emotion diary.
An AI diary for self-reflection with voice and text entries, a daily review and a view of changes over time. We built the app from scratch and released it for Android under a separate brand.
01Why a workshop needs its own product
With our own product, we also manage what happens after launch: updates, running costs and user feedback.
02What we built
An Android app with text and voice input, an AI review after closing the day and an entry history. The public site explains the product and links to Google Play and RuStore.
03An engineering decision
The current entry is saved locally as a draft. Another dictation adds recognised words to the text, while the original audio of a new entry is not stored. This preserves edits and limits sensitive data retention.
04Checking an AI feature
In one engineering example, we wrote 30 scenarios and refusal conditions before comparing models, then ran each scenario five times. Boundary cases were considered separately. This set informed model selection; it does not measure quality across all product users.
05A duplicate-reply risk caught before release
A review found that a retry after failure could call the model again, save another reply and send another notification. We limited the retry to the model request, tied the reply to its original message and separated notification delivery. With model delays and notification failures injected, the check produced one saved reply and one notification per message. This example contains no failure-rate measurements from real traffic.
06What we check after release
We track use of entries and daily reviews, collect feedback on AI responses and monitor processing cost. Errors and feedback inform the next release. These are checks on our own product, not claims about client outcomes.
07IOANN's role
The whole product: hypothesis, design, engineering, launch and iterations. It's the one case where we are both the builder and the client.
Stack
See the product
Related
Client names, metrics and screenshots are published only with written consent. In sensitive domains the product supports the decision and does not replace a specialist; responsibility and regulation remain with the client.