cito! chatbot

What it looks like

👆 Translation from Russian to English is done by Telegram, so it sounds like two pre-schoolers acting out an "at the doctor's" scenario

Idea

The initial idea behind it was to get user through Akinator-like process to figure out their disease and provide some recommendations (At some point we had a lawyer who had recommended us not to use terms like diagnosis or treatment plan for this project) they could follow before reaching for professional help. So, the bot is by no means a replacement for a doctor, but rather a tool to get a quick checkup before, but not instead of getting to a hospital.

Wrapping such a complex thing into a gamified questionnaire had hands:

• Contrary to Akinator, users don't know what they're looking for exactly.
• Most of the symptoms can be diseases by themselves, but also be a part of loads of other diseases. For a very bad example, headache can mean that user has a fever or a brain tumor. So we have to be precise with all the recommendations;
• At the same time, we can't go into too much detail, since it can be overwhelming for a user and they can miss something important.

Important side-note: I don't know shit about medicine and have to google almost every fancy-looking latin word I face, but on this project I was accompanied by two professional doctors handling all the medicine-related logic.

Flows

To implement branching and the overall question-to-question navigation, I implemented the questions using something I called Flows.

Considering this abstraction for questions:

const question = {
  id: 1337,
  text: 'Do you have a runny nose?',
  type: 'closed', // 'closed' | 'multiple-answers' | 'input'
  answers: [
    {
      id: 1338,
      text: 'Yes',
      triggers: 4
    },
    {
      id: 1339,
      text: 'No',
      triggers: null
    }
]

...And this one for users' sessions:

const session = {
  userId: 'egorushque',
  //...
  currentQuestion: 1337,
  upcomingQuestions: [80085, 42, ...]
}

A flow is just a collection of questions. Each time a user answers a question, there's a check if their answer triggers a flow. All the questions from the triggered flow are pushed to the upcomingQuestions on user's session. When the session is out of questions, we assume that we have enough data to create a recommendation.

const flow = {
  id: 4,
  questions: [
    ...
  ],
}

If you want to become an open-source superstar

You can try building a testing framework for Telegram bots.
Thinking about it now, I'm not sure what it should look like, but the fact is that there's still no such thing but a great demand for it.