My Projects

The system, using the NLP model, is engaged in clustering texts, i.e. grouping them by topic and assigning a semantic meaning to each topic.

Automatic system for monitoring cracks in the band saw and preventing its rupture.

Credit risk forecasting system for uploading data from the Credit History Bureau system

Development and implementation of the Customer Journey Map (CJM) analysis system for the Bank's customers in the context of channels. Closing the following marketing activities by integrating NLP into the source landscape – creating segmentation rules; text and visual display of user trajectories and clusters in the context of different channels; identification of isolated patterns in the trajectory, or in the use of special services / products in the context of segments and clusters; building hypotheses to identify segmentation shortcomings and their further closure.

A digital twin of power systems is a software (digital twin) that simulates a mathematical model of an engineering system (on the example of heat and water supply) and a complex of objects (boiler house, pumping station, tank, well, etc.), which includes elements of the system objects (boilers, pumps, regulators, etc.) and a hydrodynamic model of pipeline networks integrated with GIS software (for example: "Polyterm Zulu") from the source to the consumer.

Development of artificial intelligence technologies for the creation of new drugs and prediction of their properties. Self-learning computer algorithms based on Artificial Neural Networks (INNs) for repurposing known drug molecules and biologically active substances and their structural modification in order to optimize selectivity and activity Potential consumers of the product are pharmaceutical companies, including foreign developers and manufacturers of medicines. As a result of the project, the efficiency of search work in the creation of new drugs or the repurposing of existing ones will be increased. The use of neural network technologies will significantly affect the speed of bringing new drugs to the market for creating leads and, possibly, hits at the pre-experimental stage. Repurposing existing medicines is a highly effective alternative method for creating new medicines. The results of research that can be legally protected will be protected by patents, including potential patents for new applications. In the Know-How mode - focus libraries. The use of the results obtained by potential consumers is assumed on the basis of license agreements

The platform monitors and analyses the processes and outcome of electronic Election results.

ABI for Industrial Enterprises - A multifunctional hyperplatform for predictive analytics and big data processing for digitalization and optimization of industrial enterprises.

A system for evaluating vacancy texts based on criteria (attractiveness to candidates, non-discrimination based on gender or race, level of complexity and length of the text) and recommendations for improvements.

ABI Historic Figure Reconstruction is an ABI-instrument that constructs the historic person's digital self based on NLP-analysis of the person's letters and literary dialogues. The reconstructed digital twin of the historic person becomes "understandable" to the modern society by mapping the modern language into the language of the historic times of the person.

Analysis of historical context Program for identifying meaningful errors and mentions of historical events

Detect and identify anomalies in large systems and data streams. Rapid detection, evaluation, and isolation/identification of events and anomalies in complex multi-component systems and dynamic networks, such as: - Hardware and software systems – fast anomaly detection with low false alarm rates; - Cybersecurity – detection of cyberattacks with minimal latency at a given low level of false alarms; - Threats – quickly detect and prevent threats and enemy activity (gangs, terrorist groups); - Medicine and epidemiology – early detection of epidemics with a minimum level of false alarms;

Built on the basis of the iPavlov ABI platform, the intelligent decision support system (DSS) allows you to autonomously reconcile various financial documents within one organization, identifies financial leaks and gives recommendations for their elimination.

The system identifies the causes of rolling stock delays and gives recommendations to the logistics operator to eliminate them.

Automation system for the analysis of graphic documents. The system for automating the analysis of incoming documents is designed to recognize full names, dates of birth, addresses, cadastral numbers, passport numbers, dates of issue, registration and residence addresses. The system uses CV models to recognize the listed parameters and NLP to identify named entities and populate the corresponding databases in the system circuit.

Lexical analyzer Program for detecting lexical errors in English-language texts

Logic analyzer A program for detecting violations of logic or logical errors in the sequence of reasoning.

Development of a set of software, linguistic and behavioral models and methods for analyzing the information space of social media for automatic analysis of unstructured texts from social networks and the media.

Development of tools for analyzing multimodal text and transactional data ("TopicNet"). The Topic-Net project is a heterogeneous multimodal information analysis project (text, transactions, printed images, and time series) that aims to analyze the hidden dependencies within each of these data types.

Development of back-end components of a search and recommendation algorithm for processing the natural language of the prototype. NLP development task (general): Development of back-end components of a search and recommendation algorithm for natural language processing of a software solution prototype for finding and registering profitable purchases based on recommendations in Telegram communities using a bot (virtual assistant). Questions for discussion of the technical implementation of NLP search (taking into account that some of the components of the search algorithm are implemented by us, and full-text NLP search is implemented by iPavlov): -> iPavlov's NLP toolkit for natural language processing and the possibility of full-text search in messengers (databases downloaded from messengers), suitable for our task - > availability of ready-made solutions (components) and services suitable for our tasks, with the possibility of refinement and adaptation for us -> expertise and tools for working with semantic reasoning (in messengers), set criteria, exclude unnecessary and grade the refined result, work with libraries such as Python NLTK, etc. -> discussion of the specifics, order and scope of NLP development and interaction of technical specialists, taking into account the implementation and testing of NLP components of full-text search as part of our IT solution for targeted search in Telegram communities (the product and solutions are new, which will require separate improvements and settings in the process of development and testing)

Workplace Safety Monitoring is a prototype of a software package that predicts the dismissal of employees (Burn-out). Development of methods and technologies of artificial intelligence in the tasks of behavioral patterns, detection of atypical behavior, prediction of risks and threats, prevention of negative scenarios.

Sentiment Analysis Sentiment Analysis:

1. taken separately as a whole;

2. relative to the specific essence in the phrase;

3. phrases in context.

The number of classes in sentiment analysis is determined when setting a business task (usually 3-5 classes). To create a product, you need to use 1 machine learning model. This machine learning model was developed by the iPavlov team in more than 60 different versions. At the same time, some models were trained by us from scratch on a computing cluster. BERT Conversational by DeepPavlov, modified to a commercial version by the iPavlov team, classifies the sentiment of texts with high scores by metrics.

Forecasting systems in Agrotech include the following functionality: ● Price forecasting in the domestic and foreign markets for variety/crop; ● Forecasting of weather, precipitation, insolation, humidity for the selected field; ● Forecasting the occurrence of diseases; ● Forecast of the removal of useful elements by the crop, soil depletion during the cultivation of a certain crop; ● Yield forecast at current parameters; ● digital twin of crop vegetation - a digital model of crop growth taking into account the maximum possible number of external factors;

Calculation and robotic systems: ● calculation of the nutrition scheme - selection of the optimal nutrition scheme for a particular crop; ● calculation of technology efficiency - identification of technological factors affecting the cultivation of the crop; ● Increased control over agricultural operations and tracking of changes in time at each point within the contour; ● Implementation and control of irrigation by robotic sprinklers that control the volume and intensity of water supply of each sprinkler; ● Soil Scientist Robot - a robotic complex for improving soil efficiency; ● robot agronomist - a robotic complex - an assistant in agribusiness; ● Robot sorter, automated crop sorting; ● Robot analyst of applications and appeals - selection and prioritization of applications.

Applying artificial intelligence and GNSS-based systems to achieve increased efficiency at all stages of agricultural activities from soil preparation to harvesting. We have developed comprehensive IT & AI tools for forecasting, analyzing, growing and monitoring various crops with the creation of a digital twin and remote monitoring of all processes.

A Cloud-based or client-server installed NLP-Tool that provides Sentiment Analysis (Opinion Mining) of the incoming flow of social-media communication (blogs, sites, pages of social networks).

Three-class models of text sentiment analysis on streaming data (comments from social networks, forums and websites) based on the Transformer architecture. Working with pretrained models, as well as training the model from scratch on a compute cluster (BERT; BiLSTM; ELMo; XLNet).

Development of new architectures of strong machine intelligence based on reverse engineering of the principles of brain operation. Creation of working models of artificial psyche in the interests of developing new generations of robot operating systems and strong artificial intelligence

Architectures of hardware neuromorphic platforms, hardware implementation of artificial neural networks, creation of neural coprocessors based on memristor and optogenetic technologies, creation of a prototype of a neuromorphic computing device.

Automated control of the adhesion strength of the zinc coating to the steel base.

Laser gyroscope-based navigation systems integrated with GLONASS

On a continuous pickling unit, after the etching bath, the strip is washed. With a low quality of washing or an insufficient amount of inhibitors, chlorides remain on the surface of the pickled strip after washing and drying, which are provocateurs of technological corrosion. A technology needs to be developed to control residual chlorides on the etched strip surface in the production stream. "1. Type of product - pickled hot-rolled strip: 2. thickness - in the range from 1.4 to 6.4 mm, 3. width - in the range from 860 to 1670 mm, 4. maximum strip speed - up to 7 m/s, 5. temperature of rolled metal - no more than 65 ºC. 6. Measurement error of no more than 0.5 g/m²

Control of the amount of oil and the quality of oiling (uniformity of oil distribution over the surface) is carried out on the upper and lower sides of the rolled metal. Based on the results of the control, the system should provide information on the uniformity of the distribution of the oiling liquid over the surface of the rolled metal with an accuracy of +/-0.1 g/m². "1. Type of product – cold-rolled metal: 2. Strip thickness, mm: 0.25-4.0 3. Strip width, mm: 500-1850 4. Working speed, m/s: 0.5-6.0"

system of automated management of orders of Clients' bank cards

Development of an architecture and layout, a specialized "operating system" for programming robots and an avatar robot

Basic mobile control platform (BMPU) for: drones, robots, video surveillance with image processing, space monitoring systems (buoys, sensors) with elements of artificial intelligence. System-on-a-chip (SoC), integration project based on purchased complex functional (SF) IP blocks: GP CPU, interconnect, hi-speed peripherals (DDR/PCI/USB/...), etc. PlanarFET: 28nm, with subsequent adaptation for the Russian 28nm factory. Selection of requirements for the system based on information from existing laboratories and production facilities. The product is a control module on a printed circuit board, integrated directly into: an unmanned aerial (or other) vehicle. When integrated into devices, this is a system for processing: video information, audio information, other information from sensors (thermal imager, radar, spectrograph, chemical analyzer) Target applications: BPA, Robotic devices, Video cameras, Control and monitoring systems. Mobile control panels with the ability to display high-quality images in real time.

AI-powered EHR Screening rapidly detects rare diseases in children, slashes diagnosis time and boosts survival for newborns.

Services rendered to our Custom Software Development Clients normally by an hour-rate billing utilising various SD methodologies by highly qualified international team of iPavlov and AI-SYSTEMS.

Using Computer Vision and other AI-tech the product is a multi-layer personnel security shield for any construction crews working on building sites. The system alerts and notifies the security and site management in all cases when Façade Railings are not installed or violate the installation security standards.

Self-driving Truck Platform connected to Unmanned Logistics Corridors System

Digital twins of autonomous vehicles and their sensor suite embedded in a traffic and environment simulation.

Launch of the finished application in the Customer's circuit and introduction into trial operation.

1. Conclusion of a paid contract for maintenance (MOT) for at least 1 year from the launch of the system and further development of the transferred product at the Contractor's rates (consultations, training, assistance to users, providing access to the iPavlov Knowledge Base) 2. Providing free guarantees for updating and supporting the system during the validity of the maintenance contract. In case of refusal of maintenance with us or with our authorized dealer, the customer loses the warranty on our system.

a. The cost of production is not more than 100 thousand rubles and the absolute error in the z coordinate (height) is 0.2 mm. For the inspection of flat material moving at a speed of up to 12 m/s. b. The cost is up to 300 thousand rubles with an absolute measurement error of 1 μm, for the object of control moving at a speed of 2 m/s.

The system consists of up to 3 digital cameras located on the machine, and connected to a computing unit, which will be responsible for processing information from them (recognizing the type of attachments, determining the distance to the attachment and spatial position).

AI model with labeled data with presented metrics to recognize clothing items and accessories in photo images. The product provides customers with DS metrics to recognize clothing and accessories in photo images and compare the resulting images with similar products from the catalog.

Using Computer Vision and other AI-tech the product is a multi-layer personnel security shield for any construction crews working on building sites. The system alerts and notifies the security and site management in all cases when Façade Railings are not installed or violate the installation security standards.

A service that is able to highlight the tags of surrounding objects (such as: interior items, gadgets, appliances, cars, clothing, real estate, locations, etc.) in photos/videos where the Bank's client is present, with an assessment of their quality, current condition and high cost, as well as the ability to structure the resulting tags in sequence based on the development of the plot of the analyzed video sequence. The service accepts a photo or video file as input. A video file can optionally contain an audio track. The ability to determine the behavioral and consumer profiles of the client from videos/photos.Parameterization of the video stream in order to get a set of tags describing the analyzed video sequence. The ability to compare and search for similar videos of bank customers when selling a car by content. The descriptive part is in the form of a graph, which can be supplemented with details and other connections. Analysis of the video on the stage, filling in with information about the objects in it and the connections between them.

Video inspection of metal surfaces. The system is designed to detect rolling defects in real time at a rolling speed of up to 6 m/s, collect statistics and send an alarm signal to the operator in case of critical defects. Description: In the flow of the production unit, an installation is mounted, which includes cameras, illuminators, synchronization equipment, computing power and storage resources. Required: Requires software: performing primary processing of the photo array (division into suitable and doubtful, size optimization), based on a database of test images, recognition of defect classes on "doubtful" photos, comparing the recognition results with the suitability criteria for a specific order, returning the result to the operator's interface software (also included in the scope of development). It should be possible to reconfigure the system for new types of defects, change the architecture and train the algorithms used in the system.

Development of the configuration of the video inspection system for cold-rolled metal products based on an integrated computer vision system. To develop the configuration of the video inspection system for each of the above groups of rental defects. Develop a technology and a prototype of the system (Minimum viable product (MVP) is a product that has minimal but sufficient functions to satisfy the first consumers) The system must provide image quality of defects with an optical contrast (Weber) of at least 0.25 in the entire range of operating speeds of the units.

Surface Video Inspection Information System for Recognizing and Classifying Sandwich Panel Defects.

A system for predicting breaks and rejects on a Paper Machine (PM). The system is designed for: forecasting the production of defects on the PM visual (folds, wrinkles, holes) forecasting the production of defects on the PM regarding the specification of the finished product (weight, humidity, absorbency, mechanical strength indicators) issuing online recommendations to prevent the production of defects on PM-10 issuing additional recommendations in advance to improve the accuracy of forecasting; determination of the main technological parameters affecting the production of defects in the collection, storage and processing of information received from production sensors, measuring devices and information systems of the Customer; visualization of information with the ability to filter data; Log data aggregation reporting; prompt notification of technological personnel about parameter deviations and emergency situations.

A system for automatic detection, identification (recognition), archiving (including an intelligent search system) of facial images in the video data streams of all connected video surveillance cameras in real time and the output of the detected image to the workstations of operators/administrators of faces in critical information infrastructures (district heating, agriculture, food production and distribution, water supply, public health, Transport Systems, Security Services, Generation, Transmission and Distribution of Electricity, Telecommunications, Goods and Services and Financial Services, etc.).

Unlock smarter oil & gas field management with AI-powered solutions, maximising hard-to-recover reserves and boosting efficiency.

The system supported the adoption of state decisions, ensuring the automatic adoption of notifications about the start of the turnover of consumer products in the economic zones. 

An expert system for the computer-aided design of radar equipment, using artificial intelligence technologies to support technical decision-making.

Our Intelligent Waste Management System is an AI- sorting software designed to work as part of a sorting and/or recycling robot at waste processing plants. The software automatically controls the quality of sorted waste moving along the target conveyor and removes non-target contaminants from the conveyor to the appropriate bin.S

The system for analyzing the psycho-emotional state of workers - Personal identification, analysis of the psycho-emotional state of an employee, transfer of data for further processing - is in demand in large industrial and mining companies.

A computer vision system for intelligent processing of photo and video information using convolutional neural networks, deep learning and a software and hardware intelligent system for searching, localizing and classifying objects in photos and videos in real time. Neural Network Models and Software for Object Recognition in Integrated Images of Different Spectral Ranges. Methods and software for expanding training samples.

Development of equipment for an unmanned aerial vehicle for operational monitoring for various purposes using artificial intelligence technologies for control, processing and interpretation of data in real time. A complex for monitoring the natural and man-made environment based on helicopter-type UAVs for the needs of geological exploration, environmental protection, monitoring of technological infrastructure, precision farming, law enforcement and search and rescue activities

The Face Detection and Recognition System for City Services is an advanced solution designed to improve the safety and efficiency of urban infrastructure. Combining computer vision and artificial intelligence technologies, this system provides reliable face recognition in real time.

Intelligent human identification system based on artificial intelligence technologies. The system is designed for automatic recognition and identification of faces in various conditions and objects of tracking.

System for calculating the dense volume of roundwood from photographs The system is designed for: Photo recognition; Determining the length of the stack; Determining the width of the stack; Determining the height of the stack; Determination of the coefficient of full-woodness; Determination of the grade of wood (roundwood); Determination of the folded volume of wood (roundwood); Determination of the dense volume of wood (roundwood); Data Analysts; Delivery of the results of the system to the Customer's interface;

Development of a high-precision laser scanner of geometric dimensions of a stationary part

Hardware and software complex for high-performance crystallization of pharmaceutically important proteins

Hardware and software complex for non-contact temperature measurement of liquid metal and heated masses

Highly Sensitive uncooled radiation photodetectors for IR imaging

Unified database with clinical recommendations with a division for profiles. 1. A platform suitable in terms of technical requirements and capabilities for posting clinical recommendations (treatment protocols), taking into account their breakdown by the profiles of medical care provided at the Almazov Centre (creation of a unified database of treatment protocols). 2. Determination of the responsible person for placement of clinical guidelines at the selected site. 3. The ability to make updates for clinical guidelines on a regular basis by the responsible person of the Almazov Centre. 4. A system of regular notification of employees about updates to the unified database of clinical guidelines. 5. The ability to control the process of familiarizing the staff of the Almazov Centre with existing clinical recommendations, as well as with subsequent changes or updates. 6. The ability to assess the understanding and assimilation of changes or updates to clinical guidelines by the staff of the Almazov Centre.

Smart home platform for autonomous energy, flexible AI-driven power trading, and real-time balance in self-organising microenergy grids.

iPavlov Intelligent Virtual Psychologist is a smart chat-bot that assists in diagnosing user's psychological profile, assesses the user's emotional state, analyses the user's reactions and mitigates /alleviates the user's levels of anxiety and emotional distress.

Testing and scaling of the product in the internal IT-landscape of the Customer

A free, personal, live digital assistant for every citizen is a new tool for the "deroutinization" of human life and democratization of AI.

Light spots (prints) are not visible visually, such defects are poorly registered at the stage of cold rolling, and become visible only at the next stages, for example, after the galvanizing process. To identify such defects after the rolling process at the 4th and 5th class mills, it is necessary to carry out additional "rubbing" of rolled metal. It is necessary to find a way to detect this type of defect during the rolling process without reducing the productivity of the units. "1. Type of product – cold-rolled metal: 2. Strip thickness, mm: 0.25-3.2 3. Strip width, mm: 780-1650 4. Maximum speed, m/s: 30

Development of a Detection of 0.1 mm defects on moving rolled products

Development of a mathematical model of spheroidization of pearlitic structures during annealing of steels.

Development of a system for measuring the thickness of zinc coating on products (sheet, wire) in a flow

Implement a web integration service, dialog script editor, and scripting with pre-trained machine learning models. Work with speech, train your own S2T and T2S models. Automation of call centers; Provision of digital services and superservices to the population An integrated approach to solving applicants' life situations through superservices Industrial bots Fulfillment of user requests according to the terms of reference "Smart Speakers" for corporate and government systems Modern neural interfaces for ERP systems The development is based on RASA and Botfront Framework; use of BERT-based models; Implementing an integration web service, a dialog script editor, and scripting with pretrained machine learning models. Work with speech, train your own S2T and T2S models.

The "Neurolearning" method, which allows you to significantly simplify and speed up the process of teaching patients neurocontrol of the project's products using individually identified neurocontrol methods.

Neurorehabilitation Simulator for Patients After Stroke and Neurotrauma Based on Neurocontrolled Hand Exoskeleton

Functional Electrical Muscle Stimulation Device to Increase the Effectiveness of Neurorehabilitation

product that can be distributed as a non-medical device that allows patients to conduct sessions at home and combines assistive (orthosis) and rehabilitation functions

A complex for real-time functional brain mapping, which is used to assess the patient's cognitive state and create highly effective individual neuroheadsets

Electrophysiological Neurointerfacing Systems with Optogenetic Capabilitiesand Neuromorphic Processing

Discover our cutting-edge robotic telescope network—AI-powered detection, tracking and monitoring of near-Earth space objects.

AI-powered hospital terminal streamlines patient registration, queueing, access to records and payments for faster, easier medical care.

Modular system for continuous monitoring of extended pipeline sections based on deformation sensors integrated into the protective coating of pipes, and leakage points, transported liquid and gaseous products with the possibility of further expansion of the system according to other parameters of the technical condition of the pipeline.

1. Formal demonstration of the first release version of the product to the customer on the performer's servers 2. Obtaining a protocol of demonstration of the finished product indicating the volume of non-conformities, improvements, wishes (additional work) for new functional requirements and other comments with agreed deadlines and budget. 3. Signing of an additional agreement, in case of new functional requirements from the customer 4. Transfer of technical documentation for the possibility of integration and testing (approbation) of the product in the customer's internal circuit

1. Unit and functional testing; 2. Load testing; 3. Interface testing; 4. Obtaining metrics by models; 5. Product correction based on comments (tickets to Jira) from the customer's specialized employees 5. Formation of preliminary (project) technical documentation;

High Sensitivity Fire Detectors based on PbS Colloidal Quantum Dots photoconductors.