Create new product descriptions by extracting attributes from product images. Enrich these product descriptions using previous information from other products Perform a retroactive process to update the product descriptions when new attributes are found when compared to embeddings generated by Gecko (Google model), i.e., add this information to existing similar products.

Success stories

Leading global automotive companies

What if... customer claim data could be transform into a proactive fault detection system reducing attrition and claim rate?

SERVICE

Scrum and PM

INDUSTRY

Automotive

TECHNOLOGY

Sentence Transformers, UMAP,
GCP

The client

The client, aiming to cut costs, had made the decision to remove the hood temperature sensors from the design of one of the car models. As a result of implementing our solution, it was identified that in tropical markets, that model in black color was experiencing engine failures.It was determined that this was due to the accumulation of temperature, and without the sensor, the engine would shut down preventively. This allowed the client to reintroduce the sensor into the model.

Benefits

Reduce attrition and claim rate.

Proactive fault detection ensures timely resolution of issues,
boosting customer satisfaction and minimizing downtime. Realtime
data analytics drive continuous innovation, improving
product quality and reliability. Ultimately, this leads to significant
cost savings by reducing repair expenses and enhancing brand
reputation in a competitive market.

Challenge

The primary objective of this project is to further enhance our EFD system by incorporating additional variables related to vehicle models and quality
processes. By doing so, we aim to improve the accuracy and depth of fault detection, ensuring proactive resolution and maintaining customer satisfaction.

Solutions

Early fault detection was improved by incorporating variables related to vehicle models and quality processes. These variables, such as specific model features, manufacturing process details, and previous fault counts, provide context and understanding of potential problems. By incorporating these variables into machine learning models, fault detection was made more accurate and insightful.

EFD implemented relies on the identification of predefined «error clusters» coupled with advanced data segmentation using the clustering algorithms and sentence transformers analysis based on comments from customers.

Customer comments in claims are crucial for EDF. The Hugging Face API is used to deploy Sentence Transformers models, which can transform sentences into high-dimensional vectors. This allows us to measure semantic similarity and cluster-related comments. Subsequently, the UMAP technique (Uniform Manifold Approximation and Projection) is leveraged to reduce the dimensionality of vectors generated by Sentence Transformers. UMAP is a nonlinear dimensionality reduction technique that better preserves the intrinsic clustering structures within the data. This dimensionality reduction makes it easier to visualize and analyze customer comments, providing a comprehensive view of customer perceptions and opinions.

Benefits

Improving product descriptions on the online store to enhance the customer shopping experience and increase sales.

Substantial reduce the time needed to input product attributes by leveraging cutting-edge technologies.

Challenge

The process of completing product attributes was cumbersome and time-consuming. It also resulted in incomplete or error filled product descriptions, impacting the user’s shopping experience.

Solution

Using computer vision, we generated product descriptions which contain information about attributes such as color, material, dimensions, size, etc. This information is published on the ecommerce along with the product images.
The project is based on the use of AI generative models, specifically Gemini Pro Vision, to achieve three main objectives:

Google Cloud Platform (GCP) was chosen for its robust cloud computing capabilities and built-in machine learning services, providing a reliable infrastructure that is highly scalable for heavy computational tasks.

Python was selected as a high-level, interpreted language with an abundance of AI-related libraries, allowing for easy readability, quick prototyping, and sophisticated AI and machine learning projects.

Docker was utilized to automate deployment and scaling of applications, ensuring the consistent operation of the AI model across various environments.

The Gemini Pro Vision Model and Text Embedding Gecko Model were used to offer specialized functionalities in computer vision and natural language processing respectively, providing capabilities such as sophisticated analysis of visual content and representing text data as normalized numerical vectors.