AI Product Identifier

Finding the right product from an image.

Team
tfc.ai with Nico Fahrzeugteile GmbH team
Industry
Automotive Industry
Place
Germany

Together with Nico Fahrzeugteile GmbH, a highly innovative German SME, tfc.ai has developed an AI-based web application that takes an image of a product, feeds it to a deep neural network classifier and sends back the predicted product number. The images come from end-user devices like cameras and smartphones.

The solution speeds up the customer support and ordering processes. Where in the past a qualified employee with years of experience would classify the part, often taking 10 to 15 minutes to identify the exact product number, our AI solution now shows the result in less than a second! Nico Fahrzeugteile gets a competitive advantage by automating a time-consuming task.

Screenshot of the web UI of the application. Hovering over the result table will update the image of the predicted product, so users can double-check the AI result.

Solution

tfc.ai introduced the idea and guided the customer through the process of identifying a suitable first AI project, executing a Proof-of-Concept project and later implementing a production solution. Together with a customer domain expert we identified the challenges and requirements:

  • API-based solution to suit different usage scenarios and easy integration
  • On-premise hosted solution + Cloud-based on-demand training
  • Continuous learning to support growing number of identifiable parts
  • Agile project with short feedback-loops
  • Customer has the ability to autonomously retrain the model

Project Details

Industry: Automotive
Users: Customer employees in two locations in Germany.
Time Frame: Beginning 2018 (Proof-of-Concept phase) to December 2018 with additional changes afterwards
Technologies / Products used:

  • Google Cloud Platform (Compute Engine, Storage and DNS)
  • Python, PyTorch and fast.ai Deep Learning frameworks
  • Jupyter Notebook for fast network improvements
  • REST API for UI - Backend interface
  • Nvidia GeForce GPUs for network training
  • BentoML AI Application Framework
  • To get more training data we co-developed a data augmentation robot with the Factory Planning and Intralogistics team of the Chemnitz University of Technology

The project was featured in the booklet KI im Mittelstand (PDF) (“AI for SMEs”) by Lernende Systeme, Germany’s platform for Artificial Intelligence.