The Next Byte Podcast · 19 May 2026 · 17m
PhD candidate Yayati Jadhav and Professor Alireza Farimani at Carnegie Mellon University developed a self-correcting 3D printing system using a multi-agent AI framework. The system uses cameras and a vision language model to detect defects in real-time, allowing the printer to adjust parameters like flow rate and temperature mid-print. This closed-loop approach resulted in a 5X increase in peak load capacity and consistently identified failure points better than human experts. The hosts discuss how this technology could be retrofitted onto existing high-end machines to move beyond simple detection to autonomous correction.
Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
Manufacturing5X increase in peak load capacity
An AI system uses cameras to inspect 3D prints layer-by-layer and automatically adjusts printer parameters in real-time to correct defects, preventing failed prints.
“parts that they printed using this technology showed a 5X increase in peak load capacity, so much better structural integrity than one printed from, you know, the best practices that you and I were talking about.”
Speaker 1 · Hear it at 13:03
Quality & testCarnegie Mellon University multi-agent LLM frameworkConsistently identified failure points better than 14 experts
The AI model identified failure points in 3D prints more consistently than 14 human experts by analyzing visual data and adjusting print parameters mid-print.
“This model consistently identified failure points when compared to against other additive manufacturing experts. They tested against 14 other people, I think. So not like you or me, but people that like do this for a living, for example.”
Speaker 1 · Hear it at 13:03
early-stage companies
Other shows on the same ground, closest first.
Isabelle Hachette, CEO of Interspectral, explains how her company applies AI and data fusion from medical imaging to additive manufacturing process monitoring. She details their software platform, AM Explorer, which ingests multi-sensor data to provide automatic error detection and root cause analysis for metal 3D printing. Hachette discusses how these insights help companies like Volum E reduce inspection time by 90% and accelerate the transition from prototyping to certified production.
Eythor Eiriksson, founder and CEO of euler, explains how his company uses AI to analyze existing powder bed fusion camera data for real-time monitoring and quality assurance in additive manufacturing. He details the platform's ability to predict defects like spatter and warping before they cause crashes, allowing operators to intervene early. Eiriksson discusses the challenges of data access for training models, the company's hybrid cloud/on-prem deployment strategy, and the future potential of reducing reliance on post-print CT scans.
Editors Elliot Williams and Al Williams discuss a range of hardware hacks, including a 3D printed airplane duct failure due to PLA's low glass transition temperature, and a new bridge connecting RTL-SDR to Home Assistant. They review creative clock designs like a reverse sundial and a magnetic trace-line clock, alongside a volumetric display using laser-etched glass. The episode concludes with a discussion on Dan Maloney’s decision to buy a turnkey 3D printer for a specific project, and an analysis of Maya Posch’s article arguing that LLMs lack true intelligence compared to crows.
Hackaday editors Elliot Williams and Tom Nardi discuss the ESA Jupiter-bound probe's stuck antenna, a research project from the Hasso Plattner Institute that automates laser cutter kerf measurement, and the evolution of the Coco Press chocolate 3D printer into a pre-order kit. They also cover Christopher Helmke’s modular 3D-printed screw production system, tools for cleaning 3D prints, repurposing old smartphones as Linux computers, and the upcoming transition to 2D barcodes.
Hackaday editors Mike Szczys and Elliot Williams discuss several hardware hacks, including Joe Barnard’s silo-launched thrust-vectoring model rocket, which required solving pneumatic timing and sensor failures, and Graham Jessup’s 3D-printed robotic face with machine vision tracking. They also cover Daniel Riley’s solar-powered rover, which overcame steering and traction issues through iterative frame modifications, and a robotic ball bouncer using a parallel manipulator and OpenCV. The episode concludes with an analysis of smart speaker false activations and a discussion of Northrop Grumman’s Mission Extension Vehicle, a space tow truck designed to extend satellite life in geosynchronous orbit.
The summary, chapters and topics are generated from the episode’s audio by this directory, not written by The Next Byte Podcast. The episode itself belongs to its publisher.