Our Expertise and quality promise
Consulting: ML Algorithm & Technical Design
Convolutional Neural Networks, Reinforcement Learning, Apriori Algorithm or NLP? Supervised learning or unsupervised learning?
Machine Learning offers a wide range of options, so our experts will take up your specific requirements, discuss prerequisites to optimize accuracy in prediction/analysis (e.g. good light conditions for pictures) and define a technical approach to put your idea of a ML-based product or business model to practice.
Consulting: Data Strategy
Be it predictive maintenance (for a machine park) or anomly detection in a production line (based on cameras): We need we collect data, convert it, feed it into the ML-processing unit (in real-time) and process it.
We will provide you advise on big data handling, storage (if required), treatment of sensitive data (e.g. anonymization, pseudnymisation) and hardware requirements.
Image Processing, Computer Vision
We can draw on ample expertise in computer vision and processing of video streams and pictures. With a dense network of partners for computer vision, labelling and more we are your execution partner of choice to add ML to your products or automate your workflow.
Smaller projects may implement anomaly detection in a production line, which requires training of a neural network with sample pictures of anomalies. Bigger projects may require A/B testing for camera positioning and/or light conditions as well as an algorithm for selecting pictures with best characteristics for high accuracy rates.
Workflow Automation & Optimization
If a workflow follows a certain pattern or can be classified as a routine task, then automation is within reach. We will analyze available data and determine accuracy rates for workflow automation, we will suggest a data strategy to improve results.
Using predictive methods allows us to suggest optimal decisions, which optimizes the outcome of any given decision: Be it machine utilization, timing for maintenance (predictive maintenance) or energy management of electrical devices.
Natural Language Processing (NLP)
Speech-to-text, Text-to-speech, speaker recognition, Sentiment analysis: You need this in English, German, Polish or French? You need these services based on Cloud Computing or at the edge?
We will provide you an overview of the possibilities and limits for NLP-based features for your software. You have many options and we’ll help you select the appropriate one as well as implementing your ideas into your product or in your workflow.
Machine Learning and IoT
We have ample expertise with data from sources such as wearables, sensors in cars or data streams from AR glasses. At the same time we have experts on connectivity ranging from Bluetooth, 4G, Ethernet and more. We know how to minimize latency and manage huge data streams, e.g. HD video streams from AR glasses.
We will also guide you to chose the IoT Cloud Platforms that is most suitable for your needs and your budget: be it Azure IoT, Mindsphere, Teamviewer IoT or Cumulocity.
Technology Stack & Verified Expertise
ML Frameworks, Tools
Public Cloud Platforms
Developer Tools & API
ML-based real estate cost calculations
Our client company is a major player in the building trade. For each RfP they get a requirements spezification (GAEB format) between 500 and 5 000 item lines. Some items require long-standing engineering skills, others require using standard building material.
We have trained an ML-algorithm with thousands of RfP-items and building material selections. Result: Engineers can calculate projects much faster, stay easily within submitting deadlines and can optimize offerings thanks to that gain in time.
Key Application Functionality
- The Calculation Tool is on-premise
- Users will upload a spezification file in GAEB-format, that comprises 500 to 5 000 requirement items
- Based on trained ML-Algorithm the user gets an overview of requirement items that are mapped to a standard article of building material. Confidence level is displayed. User can approve or reject/overwrite proposals
- Training of ML-algorithm is updated on a regular basis with current data from the cost calculation department
Technolgy Stack & Processes
- It had been a challenge to extract data from an archived calculation tool, that had a Pervasive database. Data extraction proved difficult and required significant amount of manual data processing. However, this provided a big enough amount of training data, which provided the required accuracy of the ML-algorithm.
- Confidence building was an important aspect of the project: Transition phase ensured that proposed article selection was correct, trust-building took 4 months in total.