Use cases


Extraction of macroeconomic indicators from texts and their semantic links for sentiment analysis

Problem: 

There are progress reports and planned activities as well as projections and plans for the future. Unstructured data should lead to a more structured view - texts in tables-facts. 

texts and their semantic links for sentiment analysis

Solution: 

First, markers (indicators, quantitative and qualitative characteristics, institutions, activities, social object, etc.) and then relations between them (forecasts, implemented changes, current situation, etc.) are highlighted. Types of relations in this case get tonal color - negative, neutral, positive. 

Stack of technologies: Python, Pytorch, Sklearn



How to automate vehicle inspection in case of an accident?

Problem: 

A client in some cases has to take photos of his car using the mobile application for iOS/Android. The photos may be of a general layout, specific parts of the vehicle or damage caused by an accident. He can take the photos himself, in any place and at any time, but the insurance company has a number of quality requirements. The photo must be clear, not blurred, at the right distance from the subject, the subject must be clean.

Осмотр автомобиля.jpg

Solution: 

Assess the condition of the machine from photographs/video streams taken from different angles using mobile devices. Criteria used - light, blues, distance, contamination, etc.

Technology stack: 

Python, Java, TensorFlow, OpenCV


How to automate vehicle inspection when buying insurance?

Problem:

Potential fraudsters can deceive an insurance company when they do a photo inspection of their car on their own, for example, by replacing photos with older ones that do not yet have recent damage to the car. Thus, when buying insurance, they can report the damage they have actually sustained before buying the policy and repair it for free.

Страхование.jpg

Solution: 

detection of fraudulent photos (old, from the screen, layouts, in the photo editor, similar cars) based on image/video stream analysis.

Technology stack: 

Python, Java, TensorFlow, OpenCV


How to automate the calculation of the insurance bonus? Solution is customer scoring?

Problem:

Potential the insurance premium amount consists of 3 main parts: theft risk premium, total risk premium and damage risk premium. The amount of the damage risk premium is calculated based on the average premium for the brand/model (determined on the basis of the average expenses of the insurance company on indemnification of losses for the respective vehicle) and the "accident risk factor". which is determined individually for each insurance policy taking into account the peculiarities of the vehicle and its drivers.

Скоринг_мал.jpg

Solution: 

Calculation of individual level of accident risk (accident risk factor) on the basis of vehicle data, owners' history, accident history of the vehicle, fines, owners' data, etc.

Technology stack: 

Python, TensorFlow, Keras, XGBoost


Customer problem analysis: defaults and loan defaults 

Problem:

bank customers often do not repay consumer loans taken from the bank in the first two years when interest payments are highest. The bank has to reduce the risks of non-payment at this time as much as possible, i.e. cut off at the stage of making a decision on granting loans to insolvent clients.

Банкинг.jpg

Solution: 

Forecast of client default in the first and second year on the history of client transactions

Technology stack: 

Python, TensorFlow, Keras, Sklearn, XGBoost


Recognition and identification of signatures and stamps

Problem:


Clients at bank branches sign financial documents (contract, agreement, etc.) page by page, after which the operator puts a stamp. Then the paper document is scanned and stored in the system of storage of scanned images of documents. To exclude operator errors, the bank must be able to effectively recognize and identify signatures and seals in scanned document images.

Подпись и печать.jpg

Solution: 

Automatic detection, recognition and identification of graphic images (signatures, prints) from pages of different formats.

Technology stack: 

Python, Tesseract, TensorFlow, Keras, OpenCV, YOLOv3


Predicting customer demand at ATMs

Problem:


Clients withdraw money from ATMs unevenly. On days of mass payments the demand grows. On weekends and holidays, the demand often falls. In addition, the structure of demand is highly dependent on the currency denominations used. In order for ATMs not to stand idle and not be overflowing with money, the bank must effectively predict the dynamics of demand with aan solid horizon.

Очередь в банкомат.jpg

Solution: 

Client demand forecast with specified horizon on client transactions history

Technology stack: 

On demand


Extrapolation of the value per square meter in the real property market 

Problem:


Bank clients make deals when buying real estate. In these transactions, the banks act as intermediaries or sellers. To reduce risks when investing their reserves in real estate, the bank must effectively predict the dynamics of real estate prices.

кваритры.jpg

Solution:  

Forecast of prices per square meter with a given horizon on the history of transactions.

Technology stack: 

Python, TensorFlow, Keras, Sklearn, XGBoost, CatBoost


Car sales via online auctions 

Problem:


Managers sell customers' cars through online auctions within half an hour. During this time, they must sell the machine through dealers throughout the country at the best possible price for the customer.

bid.jpg

Solution:

Building recommendations based on the dealers and machines they sell.

Technology stack:

Python, Sklearn, Theano, Keras


Optimal location of cars in the city

Problem:


Cars go to signals coming from all over the city. To minimize the time of arrival (reduce mileage), you need to correctly place the cars on a map of the city.

город транспорт.png

Solution:  

Cars go to signals coming from all over the city. To minimize the time of arrival (reduce mileage), you need to correctly place the cars on a map of the city.

Technology stack: 

Python, NetworkX, Graphviz