我不再規劃每日行程：認識由 Gemma 驅動的 slothplan
I Stopped Planning My Day: Meet slothplan 🦥 — Powered by Gemma
開發者打造了終端機每日規劃工具 slothplan，利用 Google 開放權重 Gemma 模型理解多語言文字、擷取任務並安排時間表。Python 負責計算時段、加入休息時間及追蹤進度，資料以本機 JSON 儲存。模型可在本機執行，毋須把規劃內容傳送至外部推理 API。
This is my submission for the Hacktoberfest Open-Source AI Challenge — Week 3: Slow and Steady.
Planning my day sometimes takes more effort than actually doing the work.
Between thesis deadlines, DSA practice, urgent bugs, and personal goals, I wanted a planner that could understand my thoughts without making me fill out forms or drag calendar blocks around.
So I built slothplan — a minimalist, terminal-based daily planner that turns chaotic multilingual brain-dumps into realistic, time-blocked schedules.
Its philosophy? Slow and Steady.
What I Built
Instead of manually creating tasks, just type whatever is on your mind:
thesis chapter 3 finish karna 2 hours,
DSA 2 questions, fix urgent bug,
gym if time
slothplan uses Google's open-weight Gemma model to understand mixed English, Hinglish, and Marathi, extract tasks, estimate durations, and assign priorities.
Then Python takes over to:
- Generate conflict-free time blocks.
- Insert automatic 10-minute rest breaks.
- Track completed, skipped, and unfinished tasks.
- Generate weekly progress reports and personalized coaching suggestions.
No complicated dashboards. Just your terminal and one simple command.
How I Built It
The core principle behind slothplan is strict division of labor.
- Gemma handles language: Understands multilingual brain-dumps and converts them into structured JSON.
- Python handles logic: Validates the output, calculates time slots, inserts breaks, manages task rollover, and computes progress.
- Local JSON storage: Saves daily schedules and progress without requiring a database server.
- Fallback parser: A rule-based alternative keeps basic planning available when model inference is disabled.
I used Python, PyTorch, and Hugging Face Transformers to run the model locally. During development on Windows, I also worked through CPU memory-placement issues by explicitly configuring CPU inference.
Why Does Open Innovation Matter?
A personal planner can contain sensitive information about deadlines, health appointments, interviews, and everyday life.
Running an open-weight model locally means brain-dumps can be processed without sending them to an external inference API. It also avoids per-request API billing and gives developers greater freedom to experiment with compatible models.
For me, this project demonstrates that useful AI doesn't always require complex agent frameworks. A small language model combined with deterministic software can solve a surprisingly practical problem.
My Agent Session
I built and debugged slothplan with help from an AI coding agent, which assisted with the Python architecture, scheduler tests, prompt templates, model-download recovery, and CPU inference troubleshooting.
The Lazy Payoff
slothplan aims to turn a 15-minute planning routine into a quick command-line interaction.
Less time organizing. More time doing.
It helps you prioritize what matters, leave room for breaks, and accept that not every task needs to happen today.
Because the goal isn't to create the perfect schedule.
It's to create a schedule you can actually follow. 🦥
Prize Category
- Best Use of Gemma: Using an open-weight Gemma model for multilingual task extraction, prioritization, and weekly coaching.
Code
來源：DEV Community · dev.to