Top Artificial Intelligence (AI) Companies

• Title: Cervical cancer detection
• Client: MobileODT
• Expertise: Artificial Intelligence
• Focus Areas: Image Processing, Machine Learning, Pattern Prediction

Description

We developed deep-learning models for automated cervical cancer screening using mobile devices. We have worked on multiple predictive tasks to support medical practitioners, such as image quality detection, compliance with the medical protocol, cervical cancer detection, and cervix type identification, among others.

Challenge

• We faced multiple challenges guaranteeing data quality on the dataset.
• While we had tons of images, only a few had labels, making it difficult to train AI models with good performance. 

Solution

To support the lack of labels, we used semi-supervised learning, promoting consistency in results from minor data augmentations. 

We put the focus in doing pilots for acquiring more diverse data, with a human in the loop. 

Results

Published patents and articles:

• Bernat, Amir Shlomo, et al. “Automated monitoring of medical imaging procedures.” U.S. Patent Application No. 16/964,303. 
• Francisca Morgado, Pedro Dias, Reuven Weiser, Kelwin Fernandes, David Levitz. “Approaches to blur reduction in cervical images for automated visual evaluation”. SPIE Photonics West (2023). 

• Title: Sales Forecasts
• Client:
• Expertise: Artificial Intelligence
• Focus Areas: Machine Learning, Pattern Analysis & Recognition, Pattern Prediction

Description

Our sales forecasting machine learning algorithms are currently being deployed in two markets of one of the major beverage companies in the world. We’ve done production-level forecasts, deployed using MLOps best practices, both for On-Trade (Restaurants, Cafes, etc.) and Off-Trade (Retail). We’ve achieved significant improvement in current MAPE (Mean Average Percentage Error), particularly in the On-Trade business.

Solution

We’ve tested both SARIMAX and Prophet models. The models are retrained to avoid overfitting and to capture significant changes in consumers demands. We’ve used sales, weather, soccer games, holidays and other factors to capture consumers demand.

Results

We have deployed models for more than 250 SKUs for both businesses. It’s impossible to tackle and manage these models without proper MLOps techniques and working CI/CD pipelines.

• Title: Digital Twin IoT platform in the field of electromobility
• Client:
• Expertise: Artificial Intelligence
• Focus Areas: Pattern analysis & pattern recognition, Pattern prediction

Description

Digital Twin IoT platform in the field of electromobilityConception and development of a new Digital Twin IoT platform in the field of electromobility in the Microsoft Azure Cloud for live monitoring and control of IoT devices. Simulated IoT devices enabled test-driven development and virtual commissioning of the overall system even without real devices available. Data analyses are made possible by interactive dashboards and the connection of a data lake. The AI-based forecast of short and long-term customer requirements was prepared. A recommendation engine suggests promising locations and thus supports future business development.

• Title: Early web break detection: avoiding production downtime with AI
• Client: Kampf Schneid- und Wickeltechnik GmbH & Co. KG
• Expertise: Artificial Intelligence

Description

Thanks to early web break detection with artificial intelligence, Kampf Schneid- und Wickeltechnik customers will be able to significantly reduce downtimes and cut their error costs in the future.

Challenge

The machines from Kampf Schneid- und Wickeltechnik GmbH are often part of a production line, for example in the manufacture of plastic films from granulate to finished film. The systems from Kampf take over the cutting and winding of the films in the production process.

A tear in the film in the machines - a so-called web break - leads to production downtime in the entire production line and thus to high error costs for the customer. The web break can often be traced back to poor quality of the pre-product or incorrect machine parameters. In order to minimize costs for its customers, Kampf Schneid- und Wickeltechnik would like to be able to predict a web break automatically and at an early stage so that countermeasures can be initiated in good time.

The research and development project with codecentric aims to demonstrate the technical feasibility of this web break prediction with a proof of concept. The project is based on Kampf's in-house developed Industry 4.0 platform "the@vanced".

Solution

The machines from Kampf Schneid- und Wickeltechnik GmbH are supplied with the integrative Industry 4.0 platform "the@vanced". This platform supplies real-time data from the machine's PLC (programmable logic controller) as well as additional data from an external edge unit developed by Kampf itself. This data is used for production optimization, monitoring and historical considerations, for example in the event of complaints. It also collects production-relevant information such as the length of the end product that has already been wound. Early web break detection builds on the capabilities of this platform.

Proof of concept within a few weeks 

Thanks to codecentric's interdisciplinary expertise in machine learning, digital product development, product management and organizational enablement as well as the excellent cooperation with Kampf's software engineers, codecentric's consultants were able to create a proof of concept (PoC) within a few weeks. During development, we paid attention to sustainability, i.e. Kampf should be able to maintain the PoC with little additional effort and integrate it into the technology stack already in use.

Technology: Python backend with carefully selected algorithms 

Python backend reads the data from the "the@vanced" database, selects the relevant subset of sensor data and converts it into a defined format. By using the PCA (Principal Component Analysis) and DBSCAN (Density-based Spatial Clustering of Applications with Noise) algorithms, we can now predict an impending web break at an early stage. We were therefore able to dispense with the time-consuming and cost-intensive development of deep learning methods. Model prediction takes place in the PoC at regular intervals in a batch and can be carried out in real time in a productive environment.

Results

The project has shown that it is fundamentally possible to predict a rail break. 

We were able to build a PoC within just a few weeks thanks to our interdisciplinary expertise and close cooperation with Kampf. The existing data infrastructure of "the@vanced", which was co-developed by codecentric, was optimally utilized, but was also a prerequisite for the rapid success of the project. Sustainability was the focus during development: when selecting the algorithms, the solution with the least complexity was preferred. Attention was paid to easy integration into the existing technology landscape and preparations were already made for the customer's future enablement.

• Title: EliseAI: From lead to rental contract
• Client: EliseAI
• Expertise: Artificial Intelligence
• Focus Areas: Chatbots

Description

Vention developed a range of AI-powered web solutions for EliseAI and helped the New York-based startup achieve strong results.

Challenge

For the expansion of its platform, EliseAI was looking for a competent development partner with AI expertise.

Solution

 We took on a number of web development projects, including creating AI-powered automated responses and optimizing database performance. Our QA team also set up a new testing framework for timely bug fixing and better code quality.

Results

With the help of our solutions, the user experience was improved, customer onboarding was significantly accelerated and the conversion rate increased by 65%.