LLM-Based Automated Assembly Guide

My capstone project turns assembly PDFs into persona-aware step-by-step instructions, combining structured planning, interactive guidance, computer vision, and a polished product interface.

open live demo download capstone proposal

Homepage of the live LLM Assembly Instruction Platform showing the product framing, assembly workflow, and guided interface. — Homepage of the capstone product, designed around PDF-grounded instructions, 3D-guided assembly, and step-by-step workflow control.

Cole Mayke and the LLM Assembly Instruction Platform capstone team presenting their poster at the University of Guelph engineering showcase. — University of Guelph capstone showcase presentation with the project team, poster, and faculty advisor.

Capstone poster for the LLM Assembly Instruction Platform highlighting the workflow, computer-vision support, and measured results. — Poster summarizing the system workflow, computer-vision-assisted assembly validation, and the 121 s to 109.5 s completion-time improvement shown during the showcase.

recruiter case study

Framed the product around a simple loop: upload a manual, extract cited steps, assemble in a model-first interface, and verify progress without leaving the workflow.
Built the software/UI side around Next.js, timeline navigation, inline PDF context, chat-style clarification, and a landing experience designed to feel like a high-trust product rather than a school demo.
Connected the experience to live camera support, progress tracking, computer-vision-assisted verification, and persona-aware instruction generation for novice through expert users.

stack

Next.jsFastAPIOpenAIcomputer visionproduct prototyping

evidence

role

software/UI lead on a four-person capstone team, contributing to product definition, system architecture, and the user-facing assembly workflow.

problem

Static manuals are hard to follow, difficult to personalize, and weak at handling real-time confusion, task state, and step-by-step context.

ownership

I drove the product-facing web experience and helped shape how PDF parsing, structured planning, persona-aware guidance, and live assistance fit into one coherent flow.

outcome

Shipped a live prototype that takes PDF input and turns it into cited steps, 3D-first navigation, chat assistance, progress gating, and camera-assisted validation. In showcase testing, the system reduced assembly completion time from 121 s to 109.5 s, a 9.05% improvement, and reached an illustrative SUS score of 78/100.

artifact

Live Vercel demo, capstone proposal report, and a poster-backed showcase presentation from the University of Guelph capstone showcase.