Hardware · build

A water bottle that nags you on-device with edge AI

A decision tree running on an STM32 works out when you're actually sipping and when you've gone dry. No cloud, no phone app, no internet connection.

Can I build this?Intermediate · AI 4/5 · $20–50

Difficulty: Intermediate
AI-buildability: 4/5
Time: A weekend
Cost: $20–50

Tools

soldering ironmultimeter

Starter prompt

Based on "A water bottle that nags you on-device with edge AI" on Afterware: https://afterware.xyz/p/a-water-bottle-that-nags-you-on-device-with-edge-ai

Write an STM32 firmware project that reads an IMU over I2C and classifies sipping versus idle motion with a small hand-coded decision tree. Log each classification and trigger a buzzer or LED reminder after 60 minutes without a sip. Keep everything on-device with no network code.

Estimates based on the original write-up. How AI-buildability is scored. Building with an AI assistant? Connect it to Afterware (free MCP server).

Story

A small decision tree on an STM32 decides when you're drinking and when you've been ignoring your water. Everything runs on the board itself, so there's no cloud round trip and no phone app to babysit.

The pitch is simple: the classifier sits on the microcontroller, reads the sensor data, and nudges you when your intake lags. Nothing leaves the device. If you want a bottle that reminds you without phoning home, the core idea is a tiny model that fits in microcontroller memory and does its job locally.

Worth stealing for your own build: a decision tree is one of the cheapest ways to get useful on-device inference, and it's easy to hand-train or export to C. Pair it with whatever sensor you have on hand and a small reminder output, and you have the skeleton of a smart bottle.