ONVECTOR
DESIGN & MANUFACTURING INTERN
GREENTOWN LABS, BOSTON - 2026
INTRODUCTION
I joined Onvector as their first engineering intern. Nobody handed me a project on day one, so I found my own: machining plasma containment blocks on the CNC, then building out an AI system to speed up how the team researched materials.
MACHINING
One of my projects involved machining five polycarbonate blocks to contain plasma, each with different geometry, so I couldn't just reuse one toolpath across all five. I used Fusion 360 CAM and ran G-code on the HAAS CNC mill, then moved to the manual mill for the side ports and tapped holes the CNC program couldn't reach.
Polycarbonate melts if you run it too fast instead of cutting clean, so I dropped the spindle speed from what we'd been using and suggested we keep it there going forward. The slower speed left a noticeably cleaner edge. After machining I sanded out the tool marks and vapor polished each block to seal the surface.
AGENTIC RESEARCH
This was my own project. I proposed introducing the team to AI tooling for materials research, using Copilot Studio and Claude Code to build it. I initially built a detailed agentic family with five sub-agents, each handling a different part of the research pipeline. But once I started testing it, the scope was too loose. The agents didn't have clear enough parameters to work with, so their outputs weren't reliable.
I stepped back and went to basics. I sat down with the team and clarified which parameters actually mattered for each material category, insulator blocks needed different properties tracked than electrodes did. From there I researched different materials databases, then built a scraper to pull that data into a personal database structured around the exact parameters we cared about instead of whatever a given source happened to report. I fed that database back into the agent system, which gave it the scope it had been missing and made its outputs something the team could actually trust.
