To create a chat bot and the difficulties we encountered

439

People of all categories and age strata is particularly acute a question of career development and some of them are turning to career consultants. We decided that most of the problems people can organize and offer specific solutions without the help of a real person-career consultant. To this end, we have developed a bot-assistant — Qoshka.

 

The purpose of the project

To create a conversational bot that will help you to find a job, develop career strategies and to improve skills in the labour market.

 

The choice of technology

Chat-bot decided to do in NodeJS, since this technique provides good scalability and ease of development. For a quick start used Dialogflow — an open platform for creating conversational bots, with ready integration with Telegram, Facebook and other services. If your platform is not included in the list of integrations, there is a handy SDK.

 

When "habitat" for Qoshka no doubt was chosen as the most popular social network in the CIS "Vkontakte". In addition, we support a full web version of the bot.

 

Convenient Bot API in VC were needed to configure the integration, and besides the basic function, Facebook has a huge advantage over other platforms — user data.

 

Age, city of residence, education, work experience inthe CE data is critical for a resume, so Qoshka shamelessly uses them when filling out the profile and offers tips based on all known data.

 

User classification

When Qoshka have accumulated about a hundred conversations that allowed us to create a training sample, with which bot will be able to anticipate problems and propose their solution to the user. For classification of users decided to use a Bayesian algorithm, has prepared a training sample... and realized that this is not enough. To separate user types was high quality, it is necessary to collect thousands of examples for training and the hundreds were not enough.

 

Expert consultation

For an adequate understanding of the needs and concerns of users, we turned to a career counselor, who shared their insights based on their experience. After meeting with her we took a course on test development and guidance. Besides, based on the audience of Vkontakte, the majority of interviewees bot — pupils and students just starting their career.

 

The web version

For the web version was chosen CMF MODX, we have deployed the interface, which the HTTP API and Socket.IO is communicating with NodeJS server.

What challenges we faced

Dialogflow gave us a quick start, but began to limit adding new features. For example Vkontakte support custommnih keyboards to incorporate this into the existing system would have had to resort to a large number of "crutches" and inconvenience to the people who will manage it. So we came to the decision to create your own control panel dialogs.

 

The first version of such a panel though and does the job, but it is very inconvenient to manage, so we started to design user-friendly interface, which will satisfy all our needs.

The main task of any virtual interlocutor is the creation of a lasting dialogue that will not only entertain, but also to benefit the source.

 

Prospects of development

On the basis of Qoshka we plan to implement a complete learning platform where everyone can learn useful skills that will improve his career. Already preparing management courses microcontrollers and bath soaring.

 

Interesting development for us is the provision of hunting for companies. Qoshka will help you to choose employees and to remove from employers the headache on the search for and training of personnel.

Nobody has commented on this article yet, be the first to leave a comment
Leave comment
media__content

Have questions?

Our specialist will advise you

Leave a request