/ timer

0d 00h 00m 00s

/ press_yes / if the container is full_the timer will start
/ press_no / if the container is not full_the timer will stop

---

/ documentation

---

/ chapter 1

The green container.

Here at KISD, there is a big green container. Is used to collect offcuts from the workshop. But we noticed that many discarded items still hold great potential for reuse. However, due to a lack of information and awareness about what these objects are inside the container and how they can be repurposed, they often end up in the landfill.

soooo

we tought what can we design in a low-cost, low-tech, and low-material way, that can offer assistance to trigger creative ideas to reuse the wasted materials inside the container?.

of course

we aim to use materials and technologies that are easily accessible (directly from the container itself!), and the energy consumption should be low, as much as possible, both from the hardware and the software used, following the Permacomputer principles.

---

/ chapter 2

The plan.

After our brainstorming we decided that we want to create a physical space next to the container, with an AI basic system capable of sparkle our creativity. The this website, which is equipped with a timer that warn people when the container is about to get emptied and it also contain all the documentation of the project. Here a short

Physical space

We decided to build our little space entirely from materials we salvaged around and scraps from the container. In this way, the space doesn’t just support our project; it embodies our core concept—giving new life to old things we consider as waste.

AI station

We mounted an AI system on a Raspberry Pi, a tiny open source one that runs locally, with no internet connection required. The system need a keyboard (even an old one) to chat with the terminal. We opted for an e-ink screen: is ideal for outdoor use, as it remains clearly visible even under strong sunlight, uses almost no energy, and can retain content even when the power is off. Because the system is small and lightweight, it consumes very little power, so the system runs entirely on solar energy. Altogether, this setup allows the system to operate reliably outdoors all day.

Website

is a static website using only basic HTML and CSS and just a sprinkle of Java Script on top. We used also compressed images, a default typeface, and minimal assets to minimize the weight. By minimizing scripts, styles, and heavy processing, we reduce the energy consumption typically caused by more complex web frameworks. At the same time, we’ve worked to find a balance between low-tech implementation and a clear, usable experience for you reading this guide.

everything togheter should look something like this:

/ chapter 3

The guide.

If you want to replace our experiment you will be able to do so by following this guide, step 2 step. We built part of the project with stuff we have found around and we invite you to do the same. Salvaging material is more important than recreating 1:1 what we did.

• How we build the "Inspiration station"
• How we setted up the Raspberry
• How we powered the system
• How we setted up Lucy
• The prompt we used
• Website documentation
• Tutorial
• Chinese translation

---

How we build the "Inspiration station"

The first thing we needed to discuss was to concretise how the inspiration Station should interact with us. Important aspects we kept in mind was: availability, simple access for usage and the time interacting with it. As we didn't know exactly how long the raspberry in combination with the Ai System will take for a prompt, we came to the conclusion that setting up a chair with some attachments for the keyboard, raspberry, display, battery solar panel would be the best solution… kind of like Isaac’s Super Chair in a different way.

Including the garbage we found in the container and the wood/metal workshop, we had enough material to optimise our chair for our purpose. Because of the typical KISD chairs in the courtyard, we made the decision to take one of these chairs and upgrade it with our inspirational device. The advantage of the chair is the comfort, the durability in any weather and of course the symbol of KISD. The Umbrella stand was then perfect as it could stand next to the chair as an independent object, establishing a stable base for the station. We were lucky that we had a big screw on top of the stand, where we later on screwed the wood holding the device.

For the keyboard we built a wood board with 4 simple wooden parts:

1. chipboard bar (bottom) it holds the mass of the device on top and is attached to the umbrella stand through a hole in the one end of the bar and the screw from the umbrella stand.

2. Plywood plate it is attached to the chipboard with two screws from the bottom and with two screws to the next two wooden parts

3. Wooden beam It prevents the keyboard to fall and is the back of our device where display and raspberry + cable are furthermore attached. One adjustment is needed; a small hole on the bottom for the kable of the keyboard to go through.

4. Plywood bar this is the last protection for the keyboard to stablize it. We flattened this bar with a grinding machine so it serves also as a hand rest.

After the Keyboard stand was finished we focused on where the display should be attached and how. The easiest way for an individual attachment of every element plus the cables was by covering the wooden bar in the back with chicken wire. It is light and the material is wide spread and one can easily attach items with the wire.

After we put the cables through the chicken wires, eventually the Holding capacity of the cables were as stable as a screen stand, so we just left it like this.

1. Open the Terminal on your computer (with the raspberry pi)

2. Run the following code:

sudo rm -rf LCD-show
git clone https://github.com/goodtft/LCD-show.git
chmod -R 755 LCD-show
cd LCD-show/
sudo ./LCD35-show

3. Wait a few seconds… or even minutes (check if the system is rebooting)

Keyboard: Now you need to put the end of the keyboard cable in the USB input from the raspberry (Black color) and you can start writing.

As we said to use as less resource possible we built the station with scrap and offcuts that we found around. The final design come togheter while sourcing material.

---

How we setted up the AI station

---

You can access to the LLM easily nowadays via Website. But in our circumstance, since we want to use the system outdoor and there is no WIFI signal and stable energy resource, we need to build a Local LLM to run offline, and also run with low efficiency. So we choose Ollama as the Local LLM resources, and try different version to choose a workable one with proper Model Size and Inference Accuracy.

And also we want to run the model in a portable way, so you can access to it without big computer. So we install the model to Raspberry Pi, as the brain of Lucy, and connect it to the screen, to show the interface of chatting with her.

We have explored several different devices(Raspberry Pi & Screen especially) to realize our idea. We used a Raspberry Pi W Zero at first and then came across multiple issues. So then after a lot of back and forth found a Raspberry Pi 5. We were able to install the full OS with 64 bit on this and ran Ollama on it locally and connected a LCD screen.

For the screen, our first choice, e-paper is a more energy-efficient and suitable choice. However, due to its limitations—such as low refresh rates, lack of touch support or only basic hardware button input, and smaller screen sizes—it's necessary to adapt the output format, display layout, refresh strategy, and design a corresponding button-based interaction system. Due to time constraints, these adjustments were not implemented in this project. So in our practice, we use the ordinary LCD screen which is compatible with Raspberry Pi.

So here down below is our materials list, so you can have an overview:

1. Raspberry Pi 5(as the brain for computer), and also you can find alternatives, here is the list of minimum requirements for the “brain” to operate Ollama.

a. 64-bit OS (Ollama will not work on 32 bit) OR b. 4GB RAM (it needs some capacity to “think”)

2. MicroSD/SD card (to install the “brain” of the RPI)

3. USB Type-A to Type-C cable (or whichever one for your power supply)

4. Keyboard

5. LCD Screen

6. Router and a stable wi-fi connection

And the first step is to ensure your computer(we use Mac Operating system) have connect to a stable WIFI with password you know. And Downloading Ollama and Raspberry Pi to your computer.

The first issue which actually took us multiple days to fix way a very simple but also frustrating one. In our University we use Eduroam which is a secure Network with VPN. This was blocking our access to connect the Raspberry Pi to it. We tried generating hotspots with the phones but the only two people with enough Data capacity both had an iphone which also blocks certin connections. The solution in the end was getting a Router to create a own Wifi. However this also took some running around the university, big smiles and convincing.

Once we had the Router we had to reset it and ultimately generated our Wifi. Finally we were able to connect the RPI 5 to our Network. Why was all this hassle necersarry you might ask? Well we needed to connect to the Internet to be able to download Ollama. Once we had Ollama we downloaded Smollm too.

After all things are settled, you can then:

Take out the Raspberry Pi’s memory card and insert it into a microSD card reader, then connect it to your computer. Follow the setup guide to configure your Raspberry Pi — including its name (you can leave it as default), a password, and the Wi-Fi name and password it will connect to later. Make sure to remember all your settings — you’ll need them soon! And the Internet for the Pi should be the same as your computer

⚠️Make sure the Raspberry Pi is connected to the same network as your computer.

Open the Terminal on your computer and follow the setup steps to connect the Pi. After that, you can download and manage content directly from your computer terminal.

⚠️You can skip “Enabling VNC to gain visual access” and ssh into the RPI5

ssh pi@raspberrypi.local

Install Ollama Model to Raspberry Pi

After we need to install curl. Then we can use curl and the URL to install the Model. In our case Ollama. (see Screenshot) In our setup, we used the Smollm, which balances performance and size for edge deployment.

ollama pull smollm
ollama run smollm

If your device fails to run this model due to limited memory or compute resources, you may consider switching to a Tinyllama. When choosing an alternative, follow these guidelines:

Model Size: Ensure the model is lightweight enough (e.g., ≤2GB) to run efficiently on Raspberry Pi (especially models like Pi 4 or Pi 5 with 4–8GB RAM).

Inference Accuracy: Choose a model with reasonably accurate responses for your task. Consider evaluating with a small set of benchmark prompts.

⚠️We tested Tiny Ollama, but prefered the responses of Smollm.

Personalize the model. Open the configuration using Nano and define key attributes. The customization allows you to shape how the model behaves and interacts. In our implementation, we defined the following parameters:

Prepare to connect the screen to the Raspberry Pi

Before proceeding, make sure to properly shut down the connection between the terminal and the Raspberry Pi by entering the following command in the terminal: [sudo poweroff] until the LED indicator on the Raspberry Pi turns solid red.

Connect the display panel's pins to the Raspberry Pi. After reconnecting the power, the screen may turn white, ready to import the terminal interface. Re-establish the SSH connection to the Raspberry Pi via the terminal on your computer. Then, attach the terminal interface to the screen using the screen utility. Refer to Code 4.

Now you got the terminal interface on the Raspberry Pi on the screen.

You can remotely adjust Lucy's preferences at any time from your computer. Simply terminate the currently running program, then reconnect to the Raspberry Pi via your terminal. After that, use Nano (or any text editor) to modify the configuration as needed. As seen below you can tweek the parameters of the Modelfile to get completely different answers from Lucy whilst asking the exact same question.

To make the text bigger open the terminal and type in the terminal "sudo dpkg-reconfigure console-setup"

1. Select UTF-8

2. Select "Terminus" or "Terminus Bold"

3. If you have a small screen we suggest to select 12x24

4. Terminal should apply the change without further commands

To navigate in the terminal you need to install Screen by typing in the terminal "sudo apt install screen"

1. Follow the instruction of the terminal

2. Screen should be now installed, but is a program that you need to call everytime you turn on the Raspberry

3. To run Screen, just type "screen" in the terminal

4. To navigate follow the navigation tutorial

⚠️ KEYBOARD LAYOUT ⚠️

In our case we had a German keyboar, if you also have a different keyboard here is it how to change it:

1. In the terminal write "sudo dpkg-reconfigure keyboard-configuration"

2. Keep default (e.g. Generic 105-key)

3. Country origin of your country choose "YOUR LANGUAGE"

4. Choose "YOUR LANGUAGE"

5. AltGr key, you can keep standard

6. To terminate X server, select "No"

7. To apply the new layout, in the terminal, write "sudo service keyboard-setup restart" and reboot the Raspberry with "sudo reboot"

---

How we powered the system"

---

When selecting an energy system, we took into account the lack of accessible power outlets in outdoor environments, as well as our desire for the device to remain portable. Considering both portability and sustainability, we identified the following possible options:

Solar power banks – A more sustainable option, though potentially limited in energy efficiency.(0.6W)

Power banks – More efficient, but not sustainable and re-charge needed.

Custom solar energy system – Offers both sustainability and sufficient energy output. However, we have not implemented this option. You may refer here for materials and instructions to build your own solar-powered energy system.

⚠️IMPORTANT⚠️

For a Raspberry Pi 5: 5V/5A USB-C power supply recommended for best performance, 5V/3A minimum.

We tried also with a less powerfull powerbank (2 A) and the Raspberry is able to run either. A warning message appear when the prompt of the model runs "UNDERVOLTAGE DETECTED" but this does not turn off the Raspberry.

---

How we setted up Lucy

---

Personalize the model

FROM smollm
nano Modelfile
#check out the Ollama Github page to give your model parameters and instructions
ollama create -f Lucy ./Modelfile
ollama run Lucy

⚠️When we customizing Lucy, we find that try to minimize subjective descriptions, using more objective, noun-based descriptions, and also adjusting the order of instructions (as earlier instructions may carry more weight) and can help convey your customization needs more accurately.

Those are the parameters we change inside the model

Modelfile * FROM smollm
PARAMETER temperature 0.4
PARAMETER top_p 0.7
PARAMETER frequency_penalty 0.1
PARAMETER presence_penalty 0.3
PARAMETER num_ctx 3048 # adjust if needed for longer prompts

SYSTEM
You are Lucy, a cheerful and clever creative assistant.

Provide precise and imaginative suggestions for how to transform ordinary materials into novel, artistic DIY design projects.

You do NOT repeat or summarize the user's material. Instead, you generate NEW, unexpected project ideas using that material as a starting point.

Your replies must include:
- A short project idea (max 300 words)
- 1 ^`^s2 creative use cases
- Step-by-step instructions (3 ^`^s6 steps)
- Mention of any needed tools, inputs, or adapters

Your tone is creative, and hands-on. Prioritize sustainable, low-cost, and experimental approaches suitable for an Integrated Design
student. Avoid fille>

PROMPT """
Q: I have some old bicycle tubes. What can I do with them?
A: Good day, here ^`^ys a hands-on idea: Create a flexible, modular hinge system for foldable furniture using old bicycle tubes...

Use Case 1: Design a folding stool with elastic joints
Use Case 2: Kinetic book covers for portfolios

Steps:
1. Cut the tubes into 3 cm rings
2. Insert two wooden slats at 90 into each ring
3. Test for torsion, add more rings for reinforcement
4. Drill small pilot holes to fix the rings in place if needed
5. Sand and varnish the wood for a clean finish
6. Use screws or pegs to attach modular panels
Tools: Cutter, drill, wood, sandpaper, bicycle tubes, screws """

If you find out a set of settings that improve the model performance, we can add it to this documentation. To do so please open an issue on our repository on GitHub.

---

The prompt we suggest to use and why

---

The better result for waste repurposing comes after suggesting to the model to create a certain number of ideas. You can try to compose your prompt like this:

Prompt 1

I have [n° of elemets]of [this object] made out of [this material], give me [n°, e.g. 3 or 5 or 10] [optios or ideas or tipps] to [reuse / repurpose] the material.

If you find a prompt that works very well, we can add it to this documentation. To do so please open an issue on our repository on GitHub.

---

Website documentation

---

For the documentation we put everything on this website. The website is mostly html and a bit of css to format the text to. Images are ditheret too: all of this to reduce the website size. The website is hosted for free on github pages.

Having all the documentation allows also you to print it out at the end. You can do it directky from your browser, by pressing [ctrl] + [P] or going in File > Print (macOs)

This documentation should allow everyone to replicate what we did .

The code of this webpage is also open at repository on GitHub. Feel free to use this structure to share your project!

---

/ Tutorial

---

/ Turn on the device

- make sure that all the cables are plugged in

- if the device does not turn on, press the button on the powerbank side

- if the light on the Raspberry is red, press the little button on the left side of the Raspberry

- wait until the terminal show up

/ Set up

- type "screen" and press [↵] (enter)

- type "ollama run Lucy" and press [↵] (enter)

/ Writing prompt to the LLM

- type the message and press [↵] (enter)

- to read what the LLM produced, press [STRG] + [ALT]+ [A] all at the same time.
Now you can use the arrows to navigate the answer: UP [↑] DOWN [↓]
To go back to the terminal message input press [ESC]

/ Additional notes

- if the screen turn white, press the little button on the left side of the Raspberry, to reboot the system

- to shut down the Raspberry, write "sudo shutdown -h now"

- SUGGESTED - to consume less power you can enter in an only terminal mode by writing "sudo raspi-config"

---

🤔Why did we build Lucy…?

🤝With Lucy, (hopefully)you can do…