Terumo Cardiovascular is a medical device manufacturing and distribution company based in Ann Arbor, Michigan. Since its inception in 1921, Terumo has developed, manufactured and distributed medical devices such as tubing packs for cardiac and vascular surgery to more than 160 countries and regions.

Terumo’s tubing packs are currently assembled manually from multiple tubes and connectors, following written work instructions. A skipped step, wrong order, or inconsistent technique can compromise the tubing pack’s quality. To catch these mistakes, engineers review hours of video footage, which makes the assembly process time-consuming and potentially error prone.

Our AI Agent for Tubing Pack Assembly Quality is a web-based tool that reviews assembly video, identifies each step by comparing it with the work instructions, and flags any deviations.

A computer vision model detects parts in every frame. A neural network turns those detections into assembly actions. A state machine then checks the actions against the work instruction, and each out-of-sequence step becomes a timestamped error. Each error shows what the assembler did alongside the expected instruction.

The AI agent, powered by a large language model, explains errors in plain language and answers engineers’ questions. Engineers mark each flag as a real failure or a false positive, so a subject matter expert always makes the final decision.

A dashboard plays each video beside a color-coded timeline, so engineers can jump straight to the problematic step. A library lets engineers browse past camera footage. Each decision is saved with the corresponding video recording.

Built with Python and FastAPI, with a Streamlit dashboard, the system uses YOLO for computer vision and PyTorch for the neural network. Azure Blob Storage holds the videos, and an Azure SQL Database holds the results.