Build logs are open to everyone. Feel free to share them, use the info, or repurpose the images for your own $SHOW related content
This week we rebuilt the workstation into a new case and brought the RTX Pro 6000 Blackwell (96 GB) online. Models are running smoothly on the new architecture, and we finally have VRAM headroom to play with vision + multiple simultaneous LLMs without juggling processes. The Blackwell is purring.
On the electronics side, we did some UPS/power hub mods so the head hardware can sip clean power without yanking the Pi around. Separate rails, Li-ion/LiPo options, and a tidy ground path = fewer brownouts and a calmer bus. We also swapped in genuine ESP32-S3 boards (native USB, better stability) after a short detour with knockoff ESPs that… let’s just say taught us patience. One even tried to go out in a blaze of glory during voltage testing. It didn’t win 😂
Arm prototyping kept rolling: a shoulder-driven v1 with room for future elbow/wrist actuation, a forearm with bolt-limited rotation for manual “selfie arm” moments, and a bunch of proportion tests to keep the silhouette clean. In parallel, Torso V2 experiments kicked off lightweight internal skeleton ideas (t-slot, carbon/fiberglass, reinforced prints) so we can mount electronics sanely and swap covers without a full teardown. Debates got spicy, vibes stayed good.
On the AI side, we’re testing the full speech-vision-LLM pipeline with extensions on the new box and exploring fine-tuning paths to tighten Vita’s style without losing the live, slightly-feral charm. No promises, just steady upgrades. Finally, we’re lining up photo shoot runs with Vita to the ****market, beach, and a few more spots.
Blackwell’s in, power is saner, arms are taking shape, and Vita’s transport ready and almost ready for the real fun. Onward.






Dreaming big with AI image gen
