I stop seeing those issues after starting using Bambulab X1 3d printer, it just works 99 percent of the time. My old ender 3 v2 had those issues due to bed changing its shape during printing(thin metal changing shape in micro level due to temp and mechanical stress with 4 knobs pulling down), not sticking well, not enough fast cooling or cooling effects bed temp profile, way off even with auto level probes. Also open 3d printer with bunch of air flow is a killer for those prints. OP should try Bambulab like enclosed more accurate 3d printer from a local 3d print shop or a hackerspace after locking the design and enclosure iterations.
Recently my family member start to deal with Pectus excavatum, the xray provider and the doctor were not on the same platform and witnessed doctor doing manual calculation for it. I wished there was a way do it with web upload and get the result right a way that is error prone and accurate.
Can already see a nice web based AI app builder on the way, also can be used to quickly try apps before pushing to the device as interactive app store.
Please watch out some qemu targets for Cortex m0 or m3, increasing ram from linker out of supported ranges will cause random crashes. Would love to contribute if you are looking for people to crash some issues.
I would always prefer to start with a kit that will work right a way end to end, than focus on each problem subset to do deep dive. so101 is a highly available 3d printed and servo motor based robotic arm. That can be combined with nvidia orin like sbc as a good starting point.
Worth to check also https://www.printables.com/ for small cute 3d printed robot projects that you can build using stm32 or esp32 boards. You shall check Adafruit/Sparkfun like websites for dev boards, they even support micro python for quick prototypes.
You know that feeling when your microcontroller crashes and you spend 3 hours staring at cryptic registers trying to figure out why? Yeah, I got tired of that.
So I built an MCP server that lets Claude talk directly to GDB.
Now instead of manually decoding CFSR registers, I just ask "why did it crash?" and get back "division by zero at line 142 in calculate_average()".
It's pretty satisfying to watch Claude diagnose a deadlock between two RP2040 cores in 10 seconds - something that would've ruined my entire afternoon.
i will try Claude Code with Opus4.5 on a similar MVP, if it still fails when session becomes higher that means i need to do manual orchestration. I wonder if there is already a pattern people use for that.
From my exprience also I agree that Auto on all the time is the way to go for now,
only pay extra for initial plan nothing else. Also avoid long sessions, after first context summarization Opus4.5 is pretty much done using Cursor, will check how others do and update here.
Tried both, openai models are also good, but something with agent getting out of plan after iterating for me, initial plan says to use xyz library and it uses, later on something goes wrong and fallbacks to a hack or simpler solution despite I correct again and again. I feel like summary of summary for the context causing the issue. Do you use it as snapshot task handling or MVPs? How do you manage long sessions?
This is something I can help with - happy to do a free session to talk through onboarding strategies for your team. Feel free to reach out, contact info is in my profile.
reply