Creating a Self-Updating AI Expert Based on a Telegram Channel

Creating a Self-Updating AI Expert Based on a Telegram Channel

Creating a personal AI expert that provides in-depth, specific answers by analyzing knowledge from expert Telegram channels involves a significant amount of routine work. This routine work can be automated using special tools that are just beginning to appear. The resulting AI expert, based, for instance, on NotebookLM, can solve complex business problems in the style of selected well-known human experts and automatically "learn" from their new publications.

Our team are keenly following the use of AI as a personal business analyst and are constantly conducting our own experiments. I was particularly intrigued by the idea of turning AI into a "co-founder" capable of criticizing, asking tough and uncomfortable questions, helping to identify "blind spots," and taking on routine tasks. I was also impressed by the demonstration of how such a co-founder could be implemented by using AI to structure and analyze vast amounts of information from Telegram channels, turning them into a personal knowledge base and essentially "recruiting" favored experts.

This solidified an important realization: the true power of AI as an assistant or "co-founder" emerges when it operates on specific, current, and in-depth knowledge based not on generalized default data, but on real experience and insights often hidden in expert Telegram channels. Since our product already automated some of the routine tasks required for this, we decided to make some improvements and turn this into one of the full-fledged use cases for our system.

Turning Your AI Assistant into an AI Expert

The concept is simple: for your AI assistant (be it ChatGPT or another model) to provide valuable, specific, and applicable answers in your field, it needs the appropriate context and knowledge base. And where can one find more current and in-depth knowledge on many professional topics than in expert Telegram channels?

A manual approach involves manually exporting chat history from Telegram, followed by file conversion and uploading. To maintain an up-to-date database, this process needs to be repeated weekly. Junction Bot takes over this most labor-intensive part of the process: regular automatic collection and management of content from the very channels you designate as sources of valuable expertise.

You can:

• Collect messages from tens and hundreds of channels.
• Filter out unnecessary information, leaving only the most important content based on keywords or other criteria.
• Copy history from previous periods to create an initial knowledge base.
• Work with private sources by connecting your account.

Thus, with Junction Bot, you automatically form the "expert content" that allows you to get qualitatively different, more profound answers from AI. However, to work with tools like NotebookLM, one small detail is missing: the content needs to be exported outside of Telegram.

Toolkit

So far, we have only discussed working with content within Telegram. To solve the AI expert problem, we need to upload the content to a suitable AI tool (e.g., NotebookLM) and set up automatic updating of the knowledge base as new content is released. For this, we added functionality for automatically transferring channel content to a Google Doc with auto-updates. That is, we use Google Drive as a constantly replenished database, and NotebookLM as a research and analytical tool.

Step 1: Creating the Knowledge Base

Let's get down to business. After you have decided on the experts you want to "clone" into your team, you need to transfer their experience from Telegram to Google Drive. I will describe the actions for one channel, and you will need to repeat this operation for each channel you add to your knowledge base. For each channel, the setup needs to be done only once; then the system will work automatically.

First, use Junction Bot's channel history copying feature. In the settings, specify that only text content should be uploaded to Google Drive (it's important to note that we are currently focusing primarily on text content; media and other types can also be exported but are not needed for our scenario). Also, disable publication to the destination channel – this will speed up the process, as for our purposes it is sufficient to simply transfer all content to cloud storage without republishing it in Telegram.

The bot will create a "Junction Bot" folder on your cloud drive and, within it, a folder with the channel's name. Inside, there will be a Google Doc with texts from the channel. If the texts do not fit into one document (NotebookLM has a limit on the maximum document size), the system will create several.

Step 2: Enabling Automatic Updates

After the entire channel has been processed, go to the forwarding settings and enable the transfer of new messages from this channel. From now on, new messages published in the channel will be automatically added to the document in Google Docs. This completely automates the process of collecting new content and replaces the need to manually export history from Telegram weekly and upload new files to NotebookLM.

An important nuance: when the maximum document size in Google Docs is reached, a new document will be created. Therefore, if you use the same project in NotebookLM for a long time, it is good practice to periodically add the newly created files to your sources.

Step 3: Enjoying Your Personal AI Expert

The bot has collected and prepared for you a self-updating array of expert data in Google Docs. Now you can use this content as the basis for your AI assistant's work. This is not exactly "training" in the technical sense (like fine-tuning a model). It is more about providing the AI with a specific context and knowledge base that it will rely on when answering your questions.

Go to NotebookLM and, in the project settings, select the obtained documents from Google Docs as sources. And, of course, it is good to add documents with your internal relevant business data to the sources. Your personal expert is ready!

Wealth of Possibilities

Your AI assistant, having gained access to this specific database from expert Telegram channels and your internal documents, will be able to provide qualitatively different, more profound, and current answers. Relying on this context, it will be able to:

• Analyze trends and identify key theses, drawing on the opinions and case studies of real experts and your data.
• Answer your questions by finding specific examples, frameworks, and solutions described in the collected posts of experts you respect, tailored to your tasks.

• Example prompt: "Based on the provided articles from the channels, what main frameworks or approaches to solving [specific problem] were proposed or mentioned by experts?"

• Generate summaries and recommendations, summarizing the experience of many authors on a topic of interest to you.

• Example prompt: "Analyze the uploaded articles about [topic] and create a brief summary of key insights and practical recommendations from different experts."

• Act as a "Decision Buddy" or "Think Tank," but operating not with abstract knowledge, but with information from specific, verified sources and your internal data.

• Example prompt: "I am considering [idea/solution]. Act as a 'Decision Buddy.' Drawing on the experience of experts from the uploaded channels, ask me the toughest questions and identify potential risks or alternatives that may have been mentioned in the sources."

• Build Mind Maps and make podcasts; NotebookLM has separate buttons for this.

It is precisely this approach that allows AI to be transformed from a simple text generator using generalized knowledge into a true partner. Such an AI expert does not just rephrase publicly available information but helps to extract insights and make decisions based on the experience of professionals in your field whom you trust, and your specific data.

Conclusion

Please remember that this is experimental functionality. It is built into a rather complex and advanced product, which Junction Bot is, and some aspects of its use may not be obvious at first glance. Nevertheless, we are always ready to help. Reach out to our friendly support!

Please leave feedback on your experience with this functionality and your suggestions for its further development!^‍

‍