through the service. The internal system of the “Working Hands” project consists of server and client parts (mobile and web applications), a separate robotized telephony module, and external partner services connected to the system via API: FTS (self-employed tax calculations), Yandex.Money service (online-payments), Yandex.Maps (geolocation), Sms.ru (SMS service), Onesignal (Push-notifications), Point Bank (Financial calculations), TalkBank (Financial calculations). The system has open modular structure with APIs to integrate it in into partner services (white-label solutions). The project “Working Hands” has 2 key processes: the selection of contractors and check whether the contractor will come to the client’s site on time or not. In order to implement the business process for the selection of contractors, the “Working Hands” project managers place advertising notices on various work sites and forums. Customers and workers who call these ads meet a voice robot that implements the following steps: 1. Lead generation and client qualification via AI. 2. Making an application via AI 3. Selection of contractors for an application via AI 4. Control of answer the application and work starting with AI. 5. Work execution 6. Work Completion, calculation, financial transactions 7. Ratingscomplaints, claims. 8. Add workers to favorites. 9. Data fixation for further AI training. The key business process of the Workers Hands project is to verify that the Customer receives the contractors at their facility at the right time. The system of work with Customers and Contractors functions as follows: a phone call is made using one of the available methods for implementing such a call: - Own virtual PBX based on Asterisk - Services similar to VoxImplant or Oktell The synthesized voice welcomes the client / contractor and offers to place an order or become a contractor. The user further communicates with the system in the voice dialogue mode. The user's speech in the form of an audio stream is transmitted to a recognition server built on Yandex, Microsoft LUIS and Google technology (voice generation is performed on a similar technology stack). The recognition server turns the audio stream into text, and the text then enters a service that classifies the intents of the user. The training of the classifier of intents is conducted on the base of dialogues of contractors and partners with the operators of the “Workers Hands” project. The technological base of the intent analysis system is built on bidirectional LSTM neural networks with the attention-mechanism. According to the results of the classification of "intents", one or several models are selected, which are used for testing one or another "intent". In the case of the “Workers Hands” project, the main processor after the definition of “intents” is form-filling system. The form-filling system is built on the basis of structural representations, which can be called a frame. For example, a frame for placing an order for cleaning the territory after construction may consist of the following fields: - type of work - Address of the object - duration of work - preferred work date (date-recognition task) - target cost of work The field values are recognized by specialized analyzers based on regular expressions, dictionaries and individual classifiers. Recognized values are transmitted to the CRM-system, which forms the order. If any value was not recognized, the system sends information to the module, which is called Dialog Manager. This module is responsible for managing the process of dialogue with the user and decides the information to be requested. The Dialog Manager module uses for its work specialized configuration files in XML format, where the rules of work, sequences of dialogues and response templates are written. A graphical interface is also provided for the dialog module, which allows managers to build requestresponse sequence trees without the need to have knowledge of programming languages. This module contains a set of pre-configured dialogue templates, formed on the basis of audio recordings of communication between laborers and workers of the Workers Hands project. A number of heuristics are also being developed that improve the quality of the system, in particular, the form-filling system core technology is used to resolve anaphores and ellipsis. The additional module that is available in our system is a “chit-chat” or third-party communication module, which is also used for tasks when the bot could not answer the questions asked by users. The “chit-chat” module is based on a DSSM-like neural network with two blocks performing encoding (one for the system response and one for the current dialogue context). The system also provides for textual communication with customers and contractors through the most popular communication channels, which include Facebook, Telegram, Viber, Vkontakte, Telegram, WhatsApp. Communication can be initiated by the Workers Hands system, as well as by the contractorlaborer. A set of elements IT infrastructure based on docker containers and a system of orchestration and load control, ready for deployment in popular PaaS (Heroku, Azure, Google.AppEngine, Amazon Web Services). Ubuntu Linux operating systems.