Sendi · AI-Based Materials Failure Analysis
LLM-assisted metallurgical failure analysis on SEM fractographs and stress-strain data.
The build.
Multimodal pipeline that consumes SEM fractograph images and stress-strain curves, prompts Claude Opus 4.7 with structured tool-calls, and emits a candidate failure-mode report (fatigue, ductile, brittle, intergranular). Next.js frontend, FastAPI backend.
What I learned.
Define what 'broken' looks like before defining what 'working' looks like. The contract follows.
Whether it's a CV file or a domain entity — pick the one canonical place it lives, and shape the rest of the system around reading from it.
The interesting part is the problem framing. The stack should be the dullest viable option.
Outcome.
Shipped and running. Now iterating on the next milestone — see Now for the current roadmap.