Rong Feng

Rong Feng

Master's student @ University of Maryland

PROPS – AI-Powered Obsolete Parts Search System

Project Overview

PROPS (Potential Replacements for Obsolete Parts Search) is an AI-powered web application developed in collaboration with B. Braun Medical Inc., a global leader in medical device manufacturing. The system addresses a critical supply-chain problem: locating replacement parts for obsolete or end-of-life manufacturing equipment.

Before PROPS, engineers manually searched multiple vendor websites, catalogs, and distributors — a process that could take weeks or months, causing extended equipment downtime and production delays. PROPS transforms this workflow into a single, AI-driven search experience.

Problem Statement

  • Manufacturing equipment relies on legacy and specialized components that frequently become obsolete
  • Manual sourcing is time-consuming, inconsistent, and highly dependent on individual experience
  • Extended downtime leads to missed production targets, increased operational risk, and delays in medical product delivery

B. Braun needed a secure, scalable, and intuitive system to modernize obsolete-part sourcing without automating final procurement decisions.

Solution

We designed and implemented a web-based AI sourcing tool that allows engineers to:

  • Input known part identifiers (manufacturer, part number, description, vendor)
  • Trigger an AI-powered web search across multiple marketplaces
  • Receive ranked replacement recommendations with:
    • Vendor links
    • Pricing & availability (when available)
    • Confidence score (1–100)
    • Natural-language justification

The system acts as a decision-support tool, not an automated buyer — preserving engineering oversight and compliance requirements.

Key Features

  • AI-powered web search using OpenAI (ChatGPT o4-mini with Web Search)
  • Search-engine-style UI for fast adoption by engineers
  • Structured JSON outputs for consistent rendering
  • Confidence scoring & explanations for transparency
  • Mobile-responsive design (validated via on-site factory visit)
  • JWT-based authentication with enterprise-ready OAuth extensibility
  • No data persistence (privacy-first, IT-compliant design)

System Architecture

Frontend: React + Next.js, Tailwind CSS, Responsive component-based UI

Backend: Next.js API routes, OpenAI API integration, Secure JWT authentication (NextAuth)

AI Layer: OpenAI ChatGPT o4-mini, Real-time web search, JSON-formatted result ranking

Security: HTTPS for all requests, No exposed API keys, No storage of user or part data

Engineering & Design Decisions

  • Chose Next.js for full-stack simplicity and scalability
  • Used LLM web search instead of scraping or licensed databases to avoid IP and compliance risks
  • Ranked results instead of returning a single “best” answer to reduce false confidence
  • Designed UI after familiar search engines (Google / Octopart) to minimize training time
  • Optimized AI usage to keep token cost under $50

Impact & Outcomes

  • Reduced obsolete-part search time from days/weeks → minutes
  • Centralized sourcing intelligence into one interface
  • Improved engineering efficiency and reduced downtime risk
  • Delivered a deployable prototype aligned with enterprise IT policies
  • Validated through weekly sponsor reviews, factory floor site visit (Bethlehem, PA), and iterative demos

My Role

Software Engineer / AI Systems Developer

  • Designed AI prompting and result ranking logic
  • Implemented frontend search & results UI
  • Integrated OpenAI Web Search API
  • Contributed to system architecture & security design
  • Participated in stakeholder interviews and on-site user research