I realized I was essentially running a mental decision tree on very little sleep, so I decided to see if I could automate some of that signal processing.

What it does: CryDecoder analyzes short audio clips of a baby’s cry and classifies them into categories like hunger, discomfort/gas, tiredness, or general fussiness.

How it works: • Tech: On-device audio feature extraction paired with a lightweight ML model trained on labeled cry patterns. • Performance: Inference runs locally on the phone, which keeps latency low and avoids sending audio off-device. Results come back quickly enough to feel near real-time. • Philosophy: This isn’t meant to replace parental judgment. It’s intended as an extra data point — a sanity check when you’re tired and not sure what to try next.

The business side: The app currently uses a paid model with a preview. I’m an engineer first and still iterating on pricing and paywall placement.

I’d appreciate feedback on: 1. The technical approach and responsiveness 2. Whether the paywall timing feels reasonable for a utility like this

Thanks for taking a look.