Customer: European telecommunications holding and TV ratings.Task: Check if the TV channels show ads as per deal signed with the companies advertising and not less.Solution: The AI was watching TV channels and counting the frequency of certain adverts and TV spots. It was using complex search algorithms based on patterns (fingerprints). The software includes:
- Video file processing (automatic and manual task setting)
- Search engine, custom made
Viewer and Browser capable of editing analysis results.
Technologies: C++, OpenCV, ffmpeg, DirectShow, MS SQL, NSIS
Customer: accounting agency.Description: The customer needed a system to provide the full cycle of accounting information processing: input, checking, online publishing, delivery to clients.Our team developed a software package that runs on desktop computers as well as on mobile devices (Android and iOS platforms).The system includes:
- Update server (to generate updated content and provide it to users)
- Client applications for Windows, Android and iOS.
The system runs on Windows, Android and iOS.Technologies: C++, Java, JDBC, MySQL, Android SDK, Eclipse.
Customer: world’s leading anti-spam service provider.
Task: The Customer requested our team to enhance and support its product
Solution: The RELEX team designed and developed the low level (kernel) of the incoming e-mail analysis system. Based on the kernel was created a new email scanner. To improve the filtering quality, the following functions were implemented in the system:
- Message language auto-detection (without impairing the performance),
- Natural language parsing
- Collecting statistics based on the parser results.
These functional modules became the basis for an artificial intelligence system and containers for storing and accessing the accumulated statistics. Our team successfully resolved the task of filtering messages in Eastern languages (Chinese, Korean, Japanese etc.). The final application, besides the high email filtering quality and additional services, can process over 500 messages per second with full content analysis in the integrated target system.It also uses traditional methods of spam traffic limitation, like filtering the incoming TCP/IP traffic by the incoming IP address, which helps narrowing down the channel for unknown or suspicious traffic sources.Along with spam filtering, the system can identify messages of some specific categories, such as phishing, mail/safe-list and bounce/backscatter auto-generated.