TL;DR: I created this headtube design using AI and open source CAD and machine learning tools. I'm going to 3D print it as part of a bike I'm building.

PREAMBLE

I am a mechanical engineer by training, but have been working in software for the last 10 years. With the advances of AI tools and general availability of additive manufacturing, I decided to challenge myself to design a custom bike using 3d printed stainless steel lugs with carbon tubes. Because it's being 3d printed, I thought let me go a little nuts with the lugs.

Why steel? Because it's a hell of a lot cheaper than Ti and I decided not to care about the weight for a first project like this - just want to make sure it's strong enough for me to ride. I'm guesstimating the frame will be a chonky 7lbs.

I'll put links in the comment to a small site where you can see the headtube in 3d and the evolution of the design.

To be clear, this uses all open source CAD tools and most of the work is creating the optimization scripts that then run for hours (and hours and hours) on my 16GB M1 Macbook Air. However, I used GPT-6 Astra, Fable 5.1, and Opus 5.5 to actually design the part based on my prompting and write the optimization scripts.

So far, I have only finished the headtube, but the rest of the bike will come next.

I thought I'd share what I did to see if there are others who've done similar things and want to share notes.

PROCESS

1. Decided on custom geometry - obviously I had to make sure the bike worked. I used Astra to help with this process and create a 3d model of what the final bike would look like. Initially I was going to make this a 32" bike, but I decided to only do 1 crazy thing at a time and designed around 700c wheels
2. Once I was happy with the geo, I decided on tubes. I went with hefty bicycle tubes from Rockwood Composites. I like that they use 0, 90, and +/- 45 degree layups.
3. Started to get the headtube lug ready then realized... wait I'm going to 3d print this, why not do something weird. So I decided to try my hand at topology optimization. Topology Optimization is where you have a model create an optimal shape around some constraints. In my case it had to pass strength, fatigue, and buckling requirements with a 20% safety factor.
4. At this point I switched to Claude because I ran out of OpenAI credits. I had really good results with Claude. Cost vs quality was 🤌. I had Fable 5.1 design a topology optimization process. This used a custom finite-element code in Python to remove material and cut mass from a large solid that was the starting point. I basically froze the downtube and toptube sleeves and the bearing seats, then filled the rest of the space with metal that it could optimize around. It had to keep stress, buckling and alignment inside set limits for braking, lateral and vertical loads plus buckling.
5. (Side note: the way that the optimization works is really cool! It cuts the part up into ~450000 3d solid elements. Then it applies loads to the part and measures the stress and strain of each element in the model. Then it has each element decide how full it should be (between 0% and 100%) based on how much stress it experienced. Then it iterates again and again thousands of times until it converges on a weight. This is how it went from a 9kg chunk to a ~500g chunk)
6. A problem I ran into initially was that I didn't include a buckling condition in my part checks, and I let the agent run an excessively quick stress check, which underestimated the actual part stress. This allowed the branches to get really thin, so I added back a bunch of material and updated the optimization process to run a full stress check including buckling after each iteration epoch (batch of ~50 iterations).
7. After an optimization pass, it would run a larger check of all the forces I expected the part to experience to verify that the part still passed my limits. If a part failed, then it would return to the previous iteration and try the optimization again.
8. Once it converged (this took sometimes 16+ hours on my Mac running overnight), then it would show me the part in a viewer tool I had it create. I'd manually add/remove/or smooth material.
9. We ran through the process above several times until we converged on a shape
10. I had the agent do a final pass to smooth out the shape, add some extra material, and add some locator pins on the inside of the lug that I will use to align the carbon tubes with the lug during glue-up

My next step is to print the part in PLA to do a fit check with the carbon tubes.

I actually might do a full layup of this bike with the lugs in PLA with PVC pipe standing in for the carbon, just to figure out how I'm going to keep everything straight and get good bonding in the lugs.

If folks are interested, I'll continue posting as I continue the design and complete the build!

I just bought this bike for $180. The crankset and handlebars have been replaced, and the fork is obsolete, I mean, it’s nearly 30 years old. Everithing else is in good condition. I’m having trouble finding a suitable fork replacement; do you think this RockShox Quadra fork can be repaired?

Hey everyone
My steel fork only has these forward-facing eyelets (no side eyelets). What are the safest ways or best racks to mount panniers using this setup?