CS D-Day 2026 · KMUTT
MindSync is a journaling app that reads the emotion in what you write and in how you sound, and gently points out when the two don't match. Every entry stays encrypted on your phone.
One evening entry, start to finish: a mood, a spoken journal, and what Mochi noticed.
Recorded on an Android emulator against the analysis server in mock mode. The emulator microphone captures near silence, so the voice readings use the server’s sample transcript, not real speech recognition.
How a check-in works
Great, Good, Okay, Low or Rough. It belongs to this entry, not the whole day.
Type up to 2,000 characters, or talk for anywhere between 5 seconds and 3 minutes.
Emotion in your words and emotion in your voice, side by side. You can correct what Mochi heard.
When words and voice disagree, Mochi asks rather than tells. Your answer shapes what it suggests next.
What makes MindSync different
Each channel becomes a point on a map of feeling: pleasant to unpleasant across, calm to charged up. When the two points land more than 0.30 apart, Mochi gently mentions it. Try an entry.
Honestly, I'm fine. It was just a busy week.
Distance 0.57
Something to sit with
Your words sound steady, but your voice carries some heaviness. Sometimes we hold more than we realize. Does that feel true?
Nothing to raise
Words and voice point the same way. Mochi stays quiet.
Your companion
Mochi is not a chatbot. It never diagnoses, never gives advice, and never tells you what you really feel. It asks short questions, keeps you company while you record, and sits with you on the harder days.
No mouth, on purpose. Mochi's feelings show in its eyes and posture, so it never seems to be talking over you.
Resting
Before you check in
Listening
While you speak
Thinking
While it reads
Warm
After you journal
Present
On harder days
Asleep
When there is nothing yet
Around the check-in
Recorded on an Android emulator against the analysis server in mock mode. The emulator microphone captures near silence, so the voice readings use the server’s sample transcript, not real speech recognition.
Mochi shows what it heard as its own reading, not as your words, and you can correct it. A fix re-runs the analysis.
Reminders arrive at the time you choose, and their wording follows how your recent entries have felt.
Every day takes the colour of its average mood, worked out from that day's entries.
Mood, the emotion in your words and in your voice, and the days they disagreed, over a week, a month or longer.
Suggestions start from what you last felt, then lean toward whatever you've told it actually helped.
Mood, journaling, recording, activities and trends all work offline. Entries wait in a queue and are read once you reconnect.
On your phone
Entries, recordings and results sit in AES-256 encrypted storage, locked with a passphrase only you know.
On the server
The analysis server reads an entry, sends back the result and keeps nothing. No database, no logs of what you wrote.
In between
Entries travel over TLS 1.2 or newer, and release builds never print request bodies to a log.
Whenever you want
Export everything as JSON or CSV through your phone’s share sheet, or delete it all in one step.
Under the hood
We compared candidate models on public datasets and shipped the ones that held up. The full method is in our capstone report.
Warm on a laptop CPU, a short written entry is analysed in about 0.07 s and an 8-second voice clip, transcription included, in about 8 s. The voice benchmark uses acted speech, so real journal recordings may score lower.
Built at KMUTT as a computer science capstone