Practically speaking, Bebop is very excited to go out, and vetos are most common going home.
A plan with options might handle that, but that makes it trickier to satisfy the novelty constraint if each days plan needs to account for what was made on the previous day. Could be interesting to see what a plan with optionality looks like!
As you said, if Bebop refuses to go home, then the model has to remember the previous state, and the difficulty increases a lot. Usually, this kind of thing would be modeled with Markov rewards, using states and transition probabilities.
It is a fun problem. I really enjoy writing like this because it always gives me something worth thinking about.
Author Here: I wrote this about using numerical optimization to solve problems in my daily life. I'm interested if anyone else has done the same, and what worked for them!
Author here: I wrote this article about a failure I had applying LLM to an enterprise codebase, and ran a quick pilot study on a similar repo to understand the failures better.
I'd interested to know, are folks seeing the same things?
Something like 98 to 99% of prints worked out. When I wrote the piece, the moments of failure shaped the process and my approach, and not representative of the average print.
For the most part, my X1C + AMS set up was just plug and play, and I tried to use the Celtics print as an example of pushing that system beyond what it could do, and how struggling through that lead to specific adaptations in the process.
My focus was really getting prints out the door to hit shipping windows, so delays were felt by the customer and the reputation of my partner. Just the overall production pressure, it's a different experience than most prints I do for fun, where a failure is an opportunity to learn, and there's time to try alternatives.
As for the 0.2mm clog, I'm nearly certain it was an issue with the specific roll of filament I was using, Overature Matte PLA in the "forrest" green color they have. Other colors in that print worked. Usually PLA clogs are cooling issues, but optimizing the print profile would have been a big slow down.
All logos were supplied by the customers, and a lot of them were custom logos. The real value, IMO, was the name written in text, which is what I'm censoring out in each of the photos. There's still enough demand using alternative logos, since the "job to be done" is providing a trading card dealer with a way to show off their name/logo in pictures of their wares.
I wouldn't describe the situation as "the printer breaks down all the time". Stuff, like plates, or PFTE tube, or nozzles would wear out and break. It's not a regular occurrence, but it does happen, and it's enough of an issue when you are trying to hit a shipping window that you'll want to take steps to avoid it. From a story/narrative perspective, that's the stuff I tend to remember, not the other 99% of prints that turn out well.
The Bambu X1C is reliable, you just can't run it 24/7 and not run into issues. The most unreliable component is probably the AMS, something about that, plus matte PLA, was causing clogs where you had to take apart the AMS and fish out the broken filament. That happened to me maybe 3 times.
In those 50 orders, most of them were for 5-10 actual card stands, and throughput was enough of an issue that I bought a second printer.
As for the clogs, that's really specific to the 0.2mm nozzle, not the 0.4mm nozzle I used after the jam. Unjamming a 0.2mm is much more difficult, since the hole is so much smaller. I did try cold pulls, and several other techniques, but I do think the root cause was one roll of matte PLA filament. Sure, there was more stuff to try, but I didn't want to sink more than one night on the problem.
As for the quality of prints? I stand by the quality of prints as good examples of what can be done on consumer hardware. Consistent matte finish, even layer lines, bold, vivid logos that matched the input pics.I'm happy to design and print something people want, and my problem was never lack of demand.
This is an underrated criticism: plastics get filtered out in certain aesthetic environments: you can't really have a well decorated room with 3D printed parts. Not everyone decorates with plastic, and I've been told this several times by friends and family who were getting printed gifts.
For most 3D printing, there are a couple ways around this: sanding + painting like you mentioned, then also sanding + casting into resin or metal. For a topomap project, I experimented with acetone smoothing, but ultimately used "adaptive" layer lines that made the layers hard to see. Another "print only" option is fuzzy skin, which does a lot to obscure the shiny plastic trinket look. All of these options take extra work, though.
With the card stands, the plastic aesthetic was less of a concern, since it was a vehicle for a vendor to get their logo in every picture/video of trading cards they took for cliens. The two things that helped me make the stands feel less cheap where using plastic with a matte finish (less "shiny plastic"), and adding 2oz worth of weights, so at least when you picked up the card stand it felt heavy.
Where I lean in with my printing now is to focus on things that are so personalized that folks get over the plastic (topography maps from home areas, card stands with their name/logo), and then audiences that don't care: children and pets.
There's a couple YT creators I really like, one of them makes incredible products and sells them from the design side, the other one is running an large and automated print farm.
Two different angles on the problem, but both are using YT for at least some of their distribution problem. I'm not a big fan of 20 printers + flex-dragons sold on social media, since it feels like a low-skill hustle, but adding content creation seems like it could work. Thanks for pointing that out!
I have just about no interest in pumping out flexi-dragons, although I do have a tupperware container full of them to give to house guests and family.
The 3D printing businesses I admire are all folks who are incredibly good designers making fun and unique products that they sell it before it's ripped off. That, plus some social media content to drive organic content can work, but it would take me years of design work to get there.
I way under-estimated how long it would take to actually design something. I did a cost breakdown ahead of time on printing time + materials, but at that time the designs were simple, just text.
As things advanced, we had people ask for logos, and recreating them is really what took time.
There is still one lever here, and that was to increase the price to make that design time actually worth it. If I had to continue, that's what I would have done, but I was still losing my weekends and my free time was just more valuable.
Lots of places charge a separate fee for the design aspect like this. Printing prices will stay the same as the time + materials is consistent, so that's what you charge the client. However, since you're having to do the design part, that's where you come up with a different pricing scheme. I've been in multiple places that had similar concepts that kept things somewhat sane.
This business ran from summer '24 to late winter '25. I looked at the time for AI generation tools, but things were pretty hacky then, and the general card stand shape had a lot of features built into it for better printing and assembly. I spent an weekend trying to do 2d to 3d on a pic of my Greyhound, and never got a 3d image that didn't have a glaring problem with it.
The one real optimization here, would be a tool that converts a logo into a multi-color print. There are some solutions like HugeForge that use height maps, or hacks you can do with an svg to convert it to an STL (shape) file, but I never found one that works. As with all this generation stuff, the killer is really the details: if things don't look good and you don't have an easy way to edit it, it's never going to work for the customer. Tracing is also just one step in the process, you still have to position it on the card stand and set up the multi-color print. That said, for complex logos, SVG -> STL might make sense.
I'm convinced I could vibe code something over the weekend that takes a logo, maps it to a set of colors using some sort of segmentation, then export that as a series of STL files that can be imported into Bambu Studio (or orca slicer) and then mapped on to a card stand.
If someone is looking for a project, an end to end "make a coffee table coaster from an image" would be a great web tool (or even CLI). Especially if you could enter the number of colors (or colors you have), modify the generated traces, and and export as a single 3MF file you can import into your slicer. That's complicated, but probably do-able in a few days.
