Development of a cascade of algorithms for recognition of registration
numbers and classification of vehicles. Integration into the client's
information system. Tuning of recognition components.

Result:

• The overall accuracy index F1 measure is not lower than 98%
• Improving the quality of vehicle movement control up to 35%
• Optimization of staff work up to 30%
• Increased throughput up to 70%
• Improving the quality of access control up to 30%
• Increase the efficiency of tariffs up to 40

Company’s requests:

• Achieve a set sales plan
• Automatically classify sales by residential complex/house/apartment type.

What we have done:

• A component that predicts the probability and date of the transaction, as well as determining negative/positive factors
• Added functionality to assess the impact of additional factors on sales.

Company’s requests:

• Automatically detect the nature of the call (problems with cargo, payment, etc.)
• Identify violations in the work of operators
• Reduce repeated and conflicting calls.

What we have done:

• Integration with IP telephony systems and with the API “Yandex” for speech recognition
• Divided the calls into classes and subclasses
• Detection of stop words, triggers and tags, violations and profanity
• Detection of new categories and new requests.

The SDK for iOS and Android with API integration, developed fully by our team, reduce the time and cost of developing a solution for using in mobile applications. Now this technology is available even for those who have no experience with artificial intelligence solutions

50% [ UP TO ] reducing the cost of development 
70% [ UP TO ] reducing the time of development

Project:
To count calories in the customer's fitness application, it was necessary to make an autonomous recognition of food, fruits, vegetables and drinks, which can be integrated as an SDK.

Realization:

• A hierarchical CNN network has been created
• Optimized to work on iPhone
• Recognizes more than 1000 food dishes
• Works as an SDK via library import.

Value:
The client will receive a unique detuning from competitors and has implemented new functionality in his application.

Project:
Automatic error classification system during QA testing, with anomaly detection system.

Realization:
Using clustering algorithms, we found hidden dependencies in the logs of the testing system. Identified important mistakes for business. A sensitivity threshold was set for detecting abnormal errors during testing.

70 000+ Logs analysed 
9+ Categories revealed
Value:
QA engineers began to use their working time effectively, dealing with business errors as a priority.

Project:
Creating a smart email assistant for appointments and meetings. Users can send a request in any form to organize a meeting (online or offline), the system understands what action they are asked to perform (create, cancel, reschedule a meeting). And also extracts entities from the text: emails, names, phone numbers, dates, names of applications, companies and uses them when creating a rally.

Realization:
More than 11 neural networks are combined into one program. The service works via API and is available as a web service.

Value:
The client received a unique product on the market.

Realization:

• A REST API server was created that took a photo of a jewelry item and compared it with the downloaded catalog
• The image search was implemented through the creation of the Elasticsearch database and the preservation of the image vectors of the catalog.

18 000+ SKU in the database
<2s System’s operating time
Value:

• Reduced refund processing time from 7 minutes to 3
• Refund errors decreased by 70%
• Fraud during the refund is excluded.

Control via video surveillance
The fight detector allows you to detect automatically non-standard behavior of people.

Identification of socially dangerous actions
Hooliganism, quarrels, mugging, scuffles.

Without human involvement
The service works with multiple cameras at the same.

Our portal of cameras for automated processing and checking the condition of vehicles significantly improves the quality and speed of maintenance. The technologies developed by us automatically detect license plates, labels, vehicle condition and the presence of items.

100+ vehicles Throughput
by 80% The inspection time of vehicles is reduced


Guaranteed level of optical object recognition accuracy of at least 98%. Increasing the thoroughness of vehicle inspection by 20%

Development of a cascade of algorithms for recognizing and classifying vehicles by their types when entering paid parking. Integration into the client's information system. Tuning of recognition components.

30% Improving the quality of access control
80% Increasing throughput
30% Optimization of staff work
30% Improving the efficiency of tariffsImproving the efficiency of tariffs

Automation and acceleration of the workflow process, verification of the correctness of filling in documents and their classification.

98% Overall accuracy rate
35% Speed boost
70% Improving the quality of documents

Increased planning accuracy up to 95%
Improving the quality of planning, loading t/s, reducing idle runs and waiting time.

Optimization of delivery processes up to 30%
Automatic calculation of complex routes using graph neural networks

Optimization of staff work up to 20%
Ability to process more orders up to 50%

Project
Checking the compliance of current indicators, as functions of equipment parameters, with the values determined at the stage of analysis of historical data. Prediction of optimal parameters of the equipment operation mode based on historical data. Solving the regression problem using ML algorithms.

Value
The target indicator is to minimize the RMSE (Root Mean Square Error) - the difference between the current and optimal values up to 70%. Reduction of losses in the selection of equipment up to 15% Increase in the efficiency of equipment up to 20%