Lead your business with data insight

11,350 man-hours, 5+ years

SAP Analytics tools: Data Service, Information Design Tool, Translation Manager, IQ, BO, BI Web Intelligence.

Data warehouse: Kimball architecture, Snowflake schema.

Database platforms: Microsoft SQL, Oracle SQL, Firebird SQL, SAP SQL (Sybase).

Due to the adaptation of an hourly-rate accounting system, business reporting requirements forced our client to quickly process massive amounts of data.

We created new data warehouse, a reporting metalayer and reports based on SAP Analytics tools. Integrated various sources to the warehouse.

• The utility company has a unified reporting system for the processes managed in different systems.
• Reduced the number of reports, increased sharing capability between departments, and improved users’ data literacy​.
• Depending on the state of urgency, data in the warehouse is updated at least nightly by a variety of sources. Automated e-mail notifications about critical indicators.

Microsoft Machine Learning, SAP Manufacturing Execution

To create a test where machine learning identifies defect type and location in printed circuit boards – a test that is simple enough to be used by inexperienced operators.

Proekspert extracted the needed data from the Manufacturing Execution System and prepared complex data for a machine learning model trainer. The model was adjusted for each production item’s specifics.

• Automatic improvement of the ML model. The system compares the new model with previous. If new is “better,” then it is replaced.
• Our solution answered two questions: What is the problem? and Where is the problem?

2100 man-hours, 1+ years

U-net, OpenCV, Postgres, CouchDB

Investigate whether it is technically and commercially feasible through machine learning to obtain information about location and composition of soccer fields.

We built an application to select areas of interest, obtain a list of football fields, and view the results using a map application. A distinction is made between the fields covered with natural grass and artificial turf.

–

PyTorch, Detectron2, fast.ai, yolov5, CVAT, BokehJS

Detect, localize, and classify traffic signs based on images the client has captured beforehand.

We developed AI-powered software that helps our client to detect, localize and classify traffic signs based on images the client had captured beforehand.

• They achieved a traffic sign detection and classification tool/software which enabled them to speed up annotating traffic signs for their customer (Oslo local government) thus creating a traffic sign database for their client.
• The tool is scalable and can be expanded for use in other cities and countries.
• In the long run, this database will help plan city traffic and compiles needed data for the use of autonomous vehicles and third-party applications.