How I Used a Local LLM to Never Wash Multiple Pans Again

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

What I Built

I built the Zero-Thought Pantry Chef, a 100% local AI cooking assistant designed to eliminate culinary decision fatigue and minimize cleanup.

While most recipe apps give you complex, multi-step instructions that require a pristine kitchen and an hour of prep, this app is engineered to reward extreme laziness without sacrificing a decent meal. You input whatever random leftovers are in your fridge, and the agent processes them through a strict "Sloth Protocol." It forces the model to generate a meal requiring a maximum of one cooking vessel, limits total prep and cook time to under 20 minutes, allows for minimal quick knife work, and mandates eating directly from the pot or microwave bowl to bypass dishwashing. It intelligently assumes you have basics like salt, spices, and cooking oil, saving exhausted professionals and students from the burden of meal planning.

Demo

Here is a look at the Streamlit UI in action. When I input random staples like yam, rice, and cheese, the Sloth Chef doesn't give me a traditional casserole recipe. Instead, it tells me to chop the yam, microwave it to soften, toss in the rice and cheese with basic seasonings, and eat it straight out of the microwave bowl for a zero-effort Yam and Cheese Bowl.

Code

The entire application runs through a streamlined Python architecture using Streamlit for the frontend and Ollama for local inference.

keerthanak-tw / sloth-chef

🦥 Zero-Thought Pantry Chef

Built for Hacktoberfest 2026: "Reward Laziness" Week (International Sloth Day).

This is a local Python web app that takes whatever leftover ingredients you have in your fridge and generates a recipe requiring the absolute minimum amount of physical effort. It rigidly enforces a "Sloth Protocol": one pot only, zero knife work (tear everything by hand), and eating straight out of the cooking vessel to avoid washing dishes.

🛠️ Stack

• Language: Python
• UI Framework: Streamlit
• Model Runner: Ollama
• Model: Llama 3.2 (running 100% locally)

🚀 How to Run

1. Install Ollama and pull the model: ollama pull llama3.2
2. Install the Python requirements: pip install -r requirements.txt
3. Start the application: streamlit run app.py
4. The app will automatically open in your default web browser at http://localhost:8501.

🌍 Why Open Innovation Matters

This project relies on open-weight models (Llama 3.2) and open-source tooling (Ollama, Streamlit) to work. Running this locally…

View on GitHub

How I Built It

I developed this using Python, Streamlit, and Ollama, executing entirely on my local machine. To accelerate the boilerplate setup, I utilized GitHub Copilot in VS Code, allowing me to focus entirely on prompt engineering and local model orchestration.

The backbone of the project is Llama 3.2, an open-weight model running locally via the Ollama harness. Instead of treating the AI as a broad knowledge base, I utilized strict system prompting to heavily constrain the model's output, forcing it to act as a hyper-efficient filter against physical labor. This approach demonstrates that you don't need massive parameter counts to build highly effective, specialized agents—a localized, 3B parameter model handles instruction-following perfectly when the boundaries are clearly defined.

Why Does Open Innovation Matter?

Relying on open-weight models and open-source local execution makes this project viable in ways a closed API never could:

• Absolute Privacy: My dietary habits and pantry contents are processed locally. I am not sending my daily routines to a third-party server.
• Zero-Cost Automation: Running inference locally means zero API subscription fees or token anxiety. I can run this app every single day to figure out dinner, and it will remain completely free forever.
• True Offline Reliability: Because the entire stack lives on my machine, it works flawlessly without an internet connection. If my mobile network drops or I am entirely offline, the Llama model still executes instantly. You shouldn't need a cloud server just to figure out how to cook your leftovers.

The Lazy Payoff

This tool entirely removes the friction between being hungry and actually eating. By mathematically eliminating measuring and dishwashing while capping prep time at 20 minutes, it saves roughly 15 minutes of recipe scrolling and 15 minutes of post-meal cleanup. I used it to figure out a zero-effort yam and cheese bowl, and sticking to the single-vessel rule meant I was back to relaxing almost immediately.