SlothDrip: Zero-Touch Plant Watering Decisions with Tabular AI + Gemma

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

I built SlothDrip - an offline-first, “lazy gardener” assistant that predicts whether your plant needs watering and gives a simple one-line action.

Instead of manually checking soil every day, users enter a few quick values:

• days since last watering
• average temperature
• humidity
• soil type

The app then returns a clear verdict:

• 💧 WATER
• 😴 DO NOTHING

This saves daily effort for home gardeners and makes watering decisions faster and more consistent.

Demo

• Live App: https://slothdrip-fcc37ir38vccqekd6m7obc.streamlit.app/
• GitHub Repo: https://github.com/AslamSujah/slothdrip

Code

The full source code is available here: https://github.com/AslamSujah/slothdrip

Main components:

• app.py → Streamlit UI
• model.py → moisture prediction logic (TabPFN + fallback)
• llm.py → local Gemma/Ollama action-note generation
• db.py → SQLite logging
• requirements.txt → dependencies

How I Built It

SlothDrip is built as a local-first AI pipeline:

Tabular prediction layer (TabPFN approach)

I use tabular features to estimate watering probability from structured garden conditions.

Natural-language action layer (Gemma via Ollama)

A local LLM converts numeric prediction into a short, friendly “lazy instruction.”

Offline persistence (SQLite)

Every prediction is logged locally so users keep plant history without cloud dependence.

Simple local UI (Streamlit)

A lightweight dashboard keeps interaction minimal and beginner-friendly.

Note on deployment During deployment, I handled a TabPFN license/runtime issue by implementing an automatic fallback model (scikit-learn), so the app stays reliable in hosted environments while preserving the same user experience.

Why Does Open Innovation Matter?

Open innovation made this project possible in a practical way:

• No monthly API lock-in for core functionality
• Local-first architecture for privacy-sensitive home data
• Flexibility to combine open-weight model workflows with fallback logic
• Resilience: the app can continue working even if one model path is unavailable

A closed-only API setup would have made this more expensive and less robust for offline or low-connectivity scenarios.

My Agent Session

I built and iterated this project with AI-assisted development workflows, including deployment troubleshooting and model fallback integration. (If needed, I can update this section with a public session link/embed.)

The Lazy Payoff

SlothDrip reduces repetitive daily checking into a quick decision flow:

• input takes under a minute
• output is immediate and actionable
• logs build a long-term watering history automatically

So the “slow and steady” benefit is real: fewer routine checks, faster decisions, less mental overhead.

Prize Categories

I’m entering the following categories:

• Best Use of TabPFN
• Best Use of Gemma