Feelix 3.0 - Learn, Explore, Grow

“How can I make Feelix simple to use and easily accessible to people who need it?”

For Feelix version 3.0, I collaborated with Claude AI to continue revising the Feelix OpenSCAD code. Feelix is a 3D-printed, user-friendly Braille writer that children can use to learn Braille. This machine addresses a real need by being specifically designed to be so user-friendly that a young child can use it for educational purposes or to communicate.

By using feedback and lessons learned from CWSF 2025, I developed additional prototypes through an interactive process of designing, coding, 3D printing, testing, and refining—often repeating these steps many times until each component printed to work, as expected. Beyond creating a practical learning tool, this project raises awareness about the daily challenges that exist and are faced by visually impaired individuals.

Last year, I started this project for my cousin, Oscar. He has been blind from birth, and I wanted to create something that could potentially improve how he lives. My aunt Emily inspired me to do this project last year, but I could tell that it was far from the final prototype. However, since then, I have grown to think about how Feelix could help others, not just close family. The new question I am asking is, "How can I make Feelix V3.0 simpler to use and more easily accessible to people who need it?" This question expands to finding a way to make Feelix something that can be used in everyday life with ease.

Last year, the focus was on how to create Feelix itself. This year, I am focusing on making Feelix a viable resource for people who will benefit from it. I think that a wide variety of people would benefit from Feelix V3.0. For example, people who wish to teach others how to learn Braille step by step, a student in a school for the visually impaired learning Braille, a family member you want to communicate with, or simply someone who is curious to learn Braille themselves. I think that Feelix can help make the world a better place by bringing people together through communication and shining light on the everyday difficulties that someone who is visually impaired experiences. Feelix. Explore, Learn, Grow.

In the creation of Feelix V1.0, the question of its originality surfaced. While researching online, I found that the closest thing to Feelix V3.0 was a machine called the "Jot-a-Dot," which is not 3D-printable or something you could easily assemble yourself. As for the Perkins Braille machine, the most common Brailler, it is large and expensive, ranging from $1,500 and up. Feelix, if printed at home, would cost 35 dollars or less. As for the design process, it started with the basic concepts from last year: input prompts into Claude AI and test the code in the PrusaSlicer software. If it works, continue; if not, try again. (See procedure in photos for elaboration.) Starting with the prototype files from Feelix V2.0, I adjusted certain parts based on where the previous faults occurred. Before the adjustments were made, however, they were drawn out on paper so that I would have a main idea of how I would like Feelix V3.0 to be improved.

Originally, Feelix had many components and unstable parts. such as tape and plastic beads. Those were the first improvements. The process itself is the same as the previous one, but my role in the collaborations was far more in-depth, with me writing most of the prompts and checking in with PrusaSlicer to ensure the correct adjustments had been made. After several days and 41 initial prompts, the prototype was ready to be printed. It printed in 9 hours and on 1 try. All parts fit correctly, and it was easy to assemble. The materials used are the same as before, with only a 15% increase in PLA filament thickness. Materials: PLA from FabLab, 2 mm screws, 8 mm springs, 10 smaller elastics.

By collaborating with Claude AI and prompting it with the correct specifications, it will eventually produce what you specified after rework of the prompts and adapting Claudes terminology. Reworking the output code with Claude AI was a continuous feedback loop:

Feeback loop; Prompt Claude with new specifications, test the output code compared to my specifications, render the part in OpenSCAD, and, if acceptable, convert the part to an STL file type and import it into the PrussiaSlicer 3D printing software.

The PLA-printed parts are tested against the specification. The complete Braille machine concept, Feelix V3.0, however, has had some major improvements.

To start, the whole machine was scaled to 70% of Feelix V1.0, the original design. 70% in this case is comparable to the size of a small notebook or passport for easy access and travelability. Another big change is the new online instructable. A large incentive to the development of Feelix V3.0 was considering how someone at home could easily locate, download, reproduce, and assemble Feelix with ease.

This is why Feelix version 3's assembly uses a pressure seal and friction fit design and uses only two screws and no tape, unlike Feelix versions 1.0 and 2.0. Most pieces are friction-press-to-seal for easy assembly, like LEGO.

More on 3D printing:

At UNB FabLab, the default infill percentage is set to 15% for their PrusaSlicer Mk4s machines. To make Feelix stronger using PLA filament, it was recommended to increase the infill percentage for durability or optimaly switch to PVB.

For Feelix 3.0 the only choice was to increase the PLA infill percentage to 40% for less breakable parts and durability. This increased the print time to 9 hours, to print all parts. In the futre, I would consider printing the roller assembly with PVB.

PVB is the strongest filament and is not currently available to UNB students. The Fab Lab only uses PVB for printing their own PrusaSlicer machine parts. Filaments are available from filaments.ca, Eureka Filaments, and Bambu.

The conclusion I drew from redesigning Feelix V3.0 is that it required many adaptations and rework to the specifications and prompt attempts to get a single part of the machine output as specified.

I had to refine the AI prompts several times and redefine my expectations when prompting Claude's LLM to make sure Claude AI would correctly understand my specifications and true intention of my prompts.

This meant sometimes appropriating Claude's terminology when re-prompting the updated specifications. This change in the thought process prior to prompting a redesign reduced the errors in Claude's output code significantly.

It is a true teaching and learning, collaborative experience. The full remodel of Feelix V3.0 spanned 41 revisions to the initial code.

More importantly, I learned that AI tools can get confused with their own results, even believing that their own flawed attempts produced the correct result. It was extremely important to write out exactly what I intend for the AI to output, based on my specifications, and ask Claude to check its own work.

In this specific case, there were several versions of 'Legacy Code', or outdated code from prior revisions, remaining in the new redesigned code that would interfere with what the new code could produce.

This would present like an invisible barrier so Claude physically could not change certain things in the code and also be unaware of its error.

In future revisions, I will put aside extra time for code corrections before Feelix needs to be sent to the 3D printer. Claude's language learning abilities are continuously improving, and this year, all of the mechanisms worked on the first attempt; however, total printing reached 9 hours.

I could improve my project by performing durability testing and a child interactability test with Feelix to see if it's something a young child may enjoy using. I'm also in the process of contacting CNIB to ask for their opinion of Feelix in regard to if it would be useful in their work.

My Parents for supporting me in STEAM- I appreciate it!

My Principal at FHS for making sure the event information got to me- I wouldn’t be here without you.

My Auntie Emily for the main idea.

My 8th grade science teacher Mr. Jones- thank you for supporting me!

Project References

This is Claude.ai, where you can prompt and design your braille machine code using AI.

Claude.ai, Privacy-first AI that helps you create in confidence. Retrieved from: https://claude.ai/

This is where you can find the free OpenSCAD software that will allow you to create your STL file for PrusaSlicer.

OpenSCAD, The Programmers Solid 3D CAD Modeller, Retreived From: https://openscad.org

This is PrusaSlicer, the free software needed to visualize the OpenSCAD code by importing STL files of what you want to print on a 3D Printer.

Original Prusa 3D printers directly from Josef Prusa. Retrieved from: https://prusaslicer.net/

This is the original Perkins braille machine, it is also what my cousin Oscar uses now. It is very big and heavy to use with small hands.

Perkins Manual Brailler Blue 11 X 11 1/2in SP NR. Retrieved from: https://brailler.perkins.org/pages/perkins-brailler

Alternate References

This is the Jot-a-dot brailler, another version of a pocket-sized braille machine.

The new way to Braille is to Jot a Dot – a portable brailling device for everyone! Retrieved from: https://int.harpo.com.pl/jot-a-dot/

This is the SMART Brailler from Perkins, a smaller version of the standard braille machine.

Perkins SMART Brailler ™. Retrieved from: https://brailler.perkins.org/pages/smart-brailler

This is a website where people with disabilities and the visually impaired go to submit ideas for accessibility challenges they've encountered, whether that be opening a specific window or making a mobile phone stand. It is a great place to go for ideas if you're interested in helping with these challenges.

Design Challenges. Solutions for device ideas submitted by the community. Retrieved from: https://www.makersmakingchange.com/s/design-challenges