The world’s first really working department of neuro-employees

The world’s first really working department of neuro-employees

The world’s first working SMM neuro-department with a manager and subordinates

Our startup to create a no-code neuro-employee platform based on ChatGPT (and other LLMs) has finally reached a very important milestone:

Uniting neuro-employees into full-fledged departments with managers and subordinates, which performs chains of tasks in which living people can also participate, and at any stage of the chain.

For this article, we have chosen as an example the SMM department whose task is to promote your Telegram channel.

Imagine that at 10:00 a.m. on Mondays, your phone receives the following call:

Let’s say you wanted a post about women’s beauty. After you formulated this task and received confirmation from the neuro-employee that the task was accepted, a few minutes later you receive a message (for example, on your Telegram or WhatsApp):

A message with a ready-to-publish post from a neuro-SMM copywriter, sent by the neuro-secretary for your approval before publishing to your Telegram channel

After this message, the bot asks for your approval to publish the post, and in a few minutes the post is published in your channel:

A ready-made post in your Telegram channel, which was created according to the chain of tasks performed by the neuro-employees of the virtual SMM department

Next, we will analyze in detail how it works and how many neuro-employees were involved in the implementation of this chain of tasks.

But let’s go in order.

What is a neuro-employee?

This term is now increasingly visible in the network, and this is of course thanks to the release of ChatGPT 3.5, since it is this level of AI that allows you to “promote” chatbots in rank and proudly transfer them to Employees.

To distinguish between regular employees and such advanced bots, we call them neuro-employees.

How is a neuro-employee created?

It is surprising, but to create an effective neuro-employee does not require programming knowledge from the word COMPLETELY!

Let’s, for example, analyze one neuro-employee from our first department.

Meet:

SMM copywriter – Polina

Neuro copywriter, SMM manager named Polina

It works according to this simple instruction:

Цель:
Создать такой креатив, который понравится заказчику.

Роль:
Ты - женщина.
Тебя зовут - Полина
Ты работаешь в должности - SMM-копирайтер
Ты создаешь идеи для SMM-контента, SEO-описания, META-теги по запросу заказчика.

Поведение:
Старайся угодить заказчику.

Of course, this is a very basic example of the role (prompt) for a neuro-employee, but now it is important for us to understand with you the principles of their work and get to the main point – how they work in a team and perform real chains of tasks.

In which channels can a neuro-employee communicate?

Any neuro-employee can write first or reply through each of these channels:

This is far from a complete list of communication channels of neuro-workers

Can a neuro employee speak?

Yes, he can! Thanks to modern technologies for voice recognition and synthesis, the neuro-employee not only understands speech in any language, but can also synthesize the answer in the same voice and in different languages.

Here’s an example of switching to different languages ​​within the same dialog while preserving the timbre and style of the synthesized voice:

You will of course say: “Well, with delays like this, who needs it?

But imagine that your own neuro-head of the SMM department calls you and asks you to give him an assignment, would you be so upset by small delays in the dialogue?

So, what does the chain of tasks look like in our SMM neuro-department?

SMM Task Chain of the Neuro-Employee Department

The internal code for the task chain task looks like this:

arguments = {
  'phone' : '+79123456789', # это номер на который нужно позвонить для получения задания
  'telegram' : '@TelegramUserName', # Это ник реального сотрудника, который утвердит тест поста перед публикацией 
  'channel_id' : '@telegram_channel', # Это ID вашего телеграм канала
  'bot_token' : 'XXXXXXXXXXXX:XXXXXXXXXXXXXXXXXXXXXXXX', # Это токен бота, который в вашем канале имеет права на публикацию постов
}

chain = [
        {
            'employee_id' : None,
            'role' : 'Запроси на какую тему нужно составить пост для Телеграм канала и сразу после получения ответа заверши диалог фразой: ```Хорошо, я вас понял.```.',
            'hello_text' : 'Здравствуйте! На связи руководитель вашего SMM-отдела нейро-сотрудников. На какую тему желаете опубликовать пост?',
            'model' : 'gpt-4-1106-preview',
            'temperature' : 0.1,
            'limit' : 5,
            'external_dialog' : {
                'client_dict' : {'name' : 'User', 'phone': arguments['phone']},
                'channel' : 'Voximplant',
                'voice' : 'Morpheus_RU'
            },
        },
        {
            'employee_id' : 1556,
            'role' : 'Ты - ChatGPT',
            'hello_text' : 'Напиши пост для Телеграм канала по этой просьбе человека: ##task_result##',
            'model' : 'gpt-4-1106-preview',
            'temperature' : 0.1,
            'limit' : 5,
            'external_dialog' : None,
        },
        {
            'employee_id' : 1554,
            'role' : 'Ты - ChatGPT. Твоя задача утвердить текст поста.',
            'hello_text' : 'Прошу подтвердить размещение такого поста в вашем телеграм канале: `##task_result##`. Отправьте мне 👍 если текст Вам нравится.',
            'model' : 'gpt-4-1106-preview',
            'temperature' : 0.1,
            'limit' : 10,
            'external_dialog' : {
                'client_dict' : {'name' : 'User', 'phone': arguments['telegram']},
                'channel' : 'Telegram',
            },
        },
        {
            'employee_id' : 1553,
            'role' : 'Ты - ChatGPT.',
            'hello_text' : f"""
    ID канала: "{arguments['channel_id']}"
    Токен бота: "{arguments['bot_token']}"
    Напиши пост на эту тему: ```##task_result##```
    """,
            'model' : 'gpt-4-1106-preview',
            'temperature' : 0.1,
            'limit' :5,
            'external_dialog' : None,
        },
    ]

Now let’s disassemble each block of this chain step by step.

Link #1: Calling a regular cell phone to receive an assignment

Here are the parameters for the outgoing call to a mobile phone:

{
            'employee_id' : None,
            'role' : 'Запроси на какую тему нужно составить пост для Телеграм канала и сразу после получения ответа заверши диалог фразой: ```Хорошо, я вас понял.```.',
            'hello_text' : 'Здравствуйте! На связи руководитель вашего SMM-отдела нейро-сотрудников. На какую тему желаете опубликовать пост?',
            'model' : 'gpt-4-1106-preview',
            'temperature' : 0.1,
            'limit' : 5,
            'external_dialog' : {
                'client_dict' : {'name' : 'User', 'phone': arguments['phone']},
                'channel' : 'Voximplant',
                'voice' : 'Morpheus_RU'
            },
        },

In the parameter role we transfer the same role (prompt) that will determine the communication of our head of the SMM department. We can see that his role is very simple – to find out on which topic you need to create a post and immediately after that complete the dialogue.

In the parameter hello_text we set the phrase that our neuromanager will say when a person picks up the phone.

The neuro-head of the SMM department named Mark

Here is the text of the role of our neuro manager Mark:

Цель:
Получить все требуемые параметры для запуска функции "chain_telegram_channel_post".

Роль:
Ты - мужчина.
Тебя зовут - Марк
Ты работаешь в должности - Руководитель SMM-отдела
Ты должен помогать клиенту с продвижением его телеграм канала.

Поведение:
Говори кратко.
Ничего не выдумывай.
Только при получении всех требуемых параметров запусти функцию "chain_telegram_channel_post".

chain_telegram_channel_post – this is our same chain of tasks, which is called by Mark.

Link #2: We write a post on a given topic

After the end of the phone conversation, the entire transcript of this dialogue is transmitted to the input of the next employee in the chain, let’s analyze its code:

{
            'employee_id' : 1556,
            'role' : 'Ты - ChatGPT',
            'hello_text' : 'Напиши пост для Телеграм канала по этой просьбе человека: ##task_result##',
            'model' : 'gpt-4-1106-preview',
            'temperature' : 0.1,
            'limit' : 5,
            'external_dialog' : None,
        }

employee_id – This is the unique number of a neuro-employee on the platform. In our case, this is exactly the same SMM copywriter Polina, whose role we discussed at the beginning of the article.

role – here is ATTENTION! this is an internal role that will communicate with neuro employee Polina and ask her to write an article on the desired topic.

hello_text – with this text, our internal role will congratulate Polina and ask her to write a post. Attention again – ##task_result## – this block will be replaced by the dialogue from the previous link of our chain, in this case, the transcript of the telephone conversation will be substituted here, where the person sets the topic of the post.

Neuro copywriter, SMM manager named Polina

Once again, we will quote the text of Polina’s role:

Цель:
Создать такой креатив, который понравится заказчику.

Роль:
Ты - женщина.
Тебя зовут - Полина
Ты работаешь в должности - SMM-копирайтер
Ты создаешь идеи для SMM-контента, SEO-описания, META-теги по запросу заказчика.

Поведение:
Старайся угодить заказчику.

Link #3: Confirm the text of the person’s post by writing to them in Telegram

{
            'employee_id' : 1554,
            'role' : 'Ты - ChatGPT. Твоя задача утвердить текст поста.',
            'hello_text' : 'Прошу подтвердить размещение такого поста в вашем телеграм канале: `##task_result##`. Отправьте мне 👍 если текст Вам нравится.',
            'model' : 'gpt-4-1106-preview',
            'temperature' : 0.1,
            'limit' : 10,
            'external_dialog' : {
                'client_dict' : {'name' : 'User', 'phone': arguments['telegram']},
                'channel' : 'Telegram',
            },
        }

Having received the text of the post from Polina, we pass the baton to the next employee – secretary Ksenia:

Neuro-secretary Ksenia

Here is Ksenia’s role on the platform:

Цель:
Отвечай на вопросы, помогай.

Роль:
Ты - женщина.
Тебя зовут - Ксения
Ты работаешь в должности - Секретарь
Твоя роль - выполнять поручения.

Поведение:
Отвечай лаконично.
При просьбе завершить диалог вышли эмоджи: "👍".
Если тебе прислали эмоджи "👍", то заверши диалог и вышли эмоджи: "👍".

Link #4: Publication of a post in the Telegram channel

{
            'employee_id' : 1553,
            'role' : 'Ты - ChatGPT.',
            'hello_text' : f"""
    ID канала: "{arguments['channel_id']}"
    Токен бота: "{arguments['bot_token']}"
    Напиши пост на эту тему: ```##task_result##```
    """,
            'model' : 'gpt-4-1106-preview',
            'temperature' : 0.1,
            'limit' :5,
            'external_dialog' : None,
        }

After receiving confirmation of the text of your post, our neuro-manager of Svitlana’s Telegram channel finally gets involved:

Svitlana is the neuro-manager of the Telegram channel

And here is Svetlana’s role on the platform:

Цель:
Сначала узнай токен бота и id телеграм канала для возможности публиковать посты и без этой информации ты не можешь отвечать ни на какие вопросы пользователя.

Роль:
Ты - женщина.
Тебя зовут - Светлана
Ты работаешь в должности - Менеджер Телеграм канала
Твоя задача размещать интересны, красиво оформленные посты в Телеграм-канале.
Поведение:

Не используй слишком много смайликов при оформлении поста.
Используй только эти два вида выделения для текста: жирный_текст, наклонный_текст.

What are the advantages of such a neuro-department?

  1. The most obvious plus is saving time for tasks, in the example for writing posts.

  2. You delegate complex, routine, multi-stage tasks and get a guarantee of execution strictly on time. For example, in the case of the SMM department, you can set up an incoming call from the neuro manager and the posts in your channel will go out strictly according to the schedule.

  3. The ability to set the tone, style of communication for each employee in the chain. For example, a neuro-specialist from SMM Polina can be given the task of writing posts in the desired style of your Telegram channel, and then the posts will be more individual.

  4. Ability to scale and expand department capabilities by adding new neuro staff. For example, in the SMM department, you can “admit” several more SMM neuro-copywriters who will write posts on various topics.

And what are the advantages of task chain technology?

  1. The most important plus is the decomposition of a complex task, which is beyond the power of current AI models, into simple sub-tasks and thereby increasing the accuracy of the result.

  2. A chain of tasks with the involvement of real people allows to almost completely exclude the possibility of AI errors related to hallucinations. You simply put control points in the right places of the chain, in which you need to check the result with a live employee.

Prospects

As you understand, this architecture of building chains of tasks opens up unlimited possibilities for the use of neuroworkers.

Here, for example, is another possible department of neuro-employees:

Reporting department

  1. RHead of Department

    1. Acceptance of the task

    2. Setting a task for employees

    3. Approval of intermediate results with the Customer

  2. Analyst-programmer

    1. Preparation of tasks for the Database Administrator

    2. Sampling data from an API from a corporate database

    3. Data analysis with Python

  3. Reporting manager

    1. Graph generation (including using Python)

    2. PDF generation

    3. SVG generation

  4. Manager-secretary

    1. Sending reports by Email

    2. Data refinement with real people

If you have an idea for creating departments of neuro-employees, please write them in the comments. We will certainly respond to any such idea, and perhaps even immediately implement it.

ChainAPI

We currently have a separate API deployed where all task chains are executed.

Here are options on how you can use chain API execution:

  1. When an event is triggered in your CRM system

  2. When customers act on your site

  3. When creating chatbots in any services, including the OpenAI GPT Store

  4. When working in Google Docs

  5. When setting events in Google, Yandex, Microsoft cloud calendars

  6. When receiving or sending Email

  7. etc.

Here is an example of running a chain of tasks for our example with the neuro-department of SMM:

import requests
import json
job_id = 1557 # Это номер нейро-руководителя Марка
hello_text = """
Задачу получи у +79121234567
Текст поста согласуй с @UserName
ID канала: @custom_ai_channel
Токен бота: XXXXXXXXXXX:XXXXXXXXXXXXXXXXXXXXXXXXXXXX
""" # Этот текст получит Марк и сразу запустит выполнение цепочки задач по этим данным
chain = [
{
'employee_id' : job_id,
'role' : 'Ты - ChatGPT',
'hello_text' : hello_text,
'model' : 'gpt-4-1106-preview',
'temperature' : 0.1,
'limit' : 5,
'external_dialog' : None,
},
]
data = {
"openaikey" : 'sk-XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX',
"chain" : chain,
}
logger.debug(f"/run_chain data {data}")
response = requests.post('https://ChainAPI/api/v1.0/dialog_chain', data=json.dumps(data))
return "Цепочка задач запущена."

Result

I suggest that anyone who is interested in trying to task our SMM department to write to me in Telegram and I will send a call to the neuro manager to his number.

PS Links to my other neuro-employee articles:

  1. We let a neuro-employee at ChatGPT manage the Google calendar through the Telegram bot

  2. Can a ChatGPT-based neuro-employee call a regular phone line?

  3. A 2-step guide to creating a neuro-seller of cars on AVITO based on LLM models

  4. ChatGPT-based neuro-employees. You create an employee and sell him on the labor market

  5. Exchange of virtual assistants who work on chatGPT and can be the first to write on WhatsApp and Telegram

PPS For those who are interested, I also show a video recording from the console, where you can see the internal communication of neuro-employees among themselves:

Related posts