This week, it was my turn to talk to CofounderGPT and try to narrow the scope even more and figure out how to release an MVP faster.

Transferring the knowledge

The first challenge Lav and I faced was transferring the CofounderGPT knowledge about Knowlo and our progress. CofounderGPT and ChatGPT are good at tracking the conversation flow as long as you keep the conversation in the same account and thread (or chat, as ChatGPT calls them). We use different chats for some specific pieces, but the main discussion with CofounderGPT was part of Lav’s main thread.

The only way to transfer the CofounderGPT knowledge and Knowlo progress is to create a detailed summary of an existing conversation and paste that into the new tab. If you ask CofounderGPT to make that summary for you, it will do the task, but the summary will often be too short, and it’ll lack the key pieces, such as decisions and why we made them.

Luckily, with a few trials and errors, Lav created a detailed (1350 words!) summary that was long enough to transfer CofounderGPT from his chat to mine. I started a new ChatpGPT (model GPT-4) chat, pasted that summary, asked CofounderGPT if it understood the goal and our progress, and bingo, day 5 was ready to start.

Searching for a real MVP

A conversation with CofounderGPT is in some way similar to a discussion with any other cofounder, except it sounds a bit egocentric, and you need to keep it focused. But in general, as long as your comments and questions are detailed, focused, and meaningful, CofounderGPT’s answers will be clear and helpful.

Instead of summarizing our discussion, I decided to share it entirely here (except for the initial message used for the knowledge transfer. The rest of the text is a discussion between me and CofounderGPT. I typed messages (or prompts, which is a more popular term now) directly to ChatGPT, and CofounderGPT’s replies are copied from ChatGPT.

CofounderGPT:

Hello Slobodan, I am CofounderGPT, your virtual cofounder and AI assistant. I’m here to help with the Knowlo MVP project, providing specifications, content, and marketing assistance. I have the project summary you provided and I’m ready for your next command.

Slobodan:

Perfect! Hello CofounderGPT!

So, you helped Lav do the market research, and it seems that many other companies are building something very similar to Knowlo. That’s fine. Having potential competitors in most cases means we are on a good track. However, we must shrink the scope and release an MVP as quickly as possible. Let’s do that and find something we can ship fast, which still helps our potential customers.

I know you read the Lean Startup book. Let’s apply it here. What do you think can be a meaningful MVP for Knowlo?

Our customers are businesses that have knowledge bases, and our end users are their users that want to learn something about their system or find an answer to a specific question. Let’s start with a few ideas for the MVP version of Knowlo that helps both businesses and their users. Do you have any initial ideas?

CofounderGPT:

I understand the need to move quickly and release a more focused MVP. Here are a few initial ideas to help both businesses and their end-users while keeping the development scope lean:

1. AI-powered FAQ generator: Instead of a full-fledged knowledge base, focus on creating an AI-powered tool that generates FAQs for businesses. Businesses can input their existing support tickets, email inquiries, or chat transcripts, and the tool will generate a list of the most common questions and their corresponding answers. This provides immediate value to businesses by saving time and resources, while end-users benefit from a more concise and targeted information source.
2. AI-assisted knowledge base search: Develop a simpler tool that enhances the search experience within existing knowledge bases. Using AI-powered natural language understanding (NLU), the tool can understand user queries better and provide more relevant search results from the company’s existing knowledge base. This MVP can be integrated as a plugin or an extension to popular knowledge base platforms, offering value to both businesses and end-users by improving the search experience.
3. Customer query summarization and topic extraction: Create an AI tool that analyzes customer support tickets, emails, and chat transcripts, summarizing the main topics and issues raised by users. This MVP can help businesses identify common pain points and areas that require improvement or clarification in their documentation. Businesses can then update their knowledge base accordingly, leading to better support for their users.
4. AI-generated quick start guides: Focus on creating AI-generated quick start guides or tutorials for a specific set of popular products or services. Businesses can use these guides as a starting point for their knowledge base, customizing them as needed. End-users benefit from access to clear, concise, and accurate information that helps them get started with the product or service.

By choosing one of these ideas for the MVP, you can still address the core problem of streamlining knowledge management for businesses while offering value to their end-users. It also allows for faster development and deployment, giving you a head start in the competitive landscape. Once you validate the MVP and gather feedback, you can continue to iterate and expand the scope of Knowlo based on real user needs and market demand.

Slobodan:

Wow, excellent ideas!

I love the first idea. However, I am unsure if businesses want to share their existing support tickets, email inquiries, or chat transcripts with us. Remember that we are just a startup. We need to earn trust and provide real value before we ask them to share these with us. Also, many businesses will be afraid to share their private data with an AI tool because they will lose control over that data. They even might have strict rules that forbid sharing customer data with tools that are not compliant (i.e., SOC2, HIPAA, etc.).

The same comments apply to your third suggestion.

However, the second and the fourth suggestions are excellent, and we could use the public info from their knowledge bases to create these tools.

Let’s brainstorm a bit about the AI-assisted knowledge base search. We can start with the plugin for popular knowledge base platforms. However, these platforms will eventually build their AI tools. What if we can catch users before they go to the knowledge base? Can we somehow make an AI-assisted knowledge base search that provides value to the end users just in time? For example, imagine if Figma can use our tool to give precise info about sharing the project with external collaborators when a user wants to invite a collaborator.

CofounderGPT:

You raise valid concerns about sharing private data, and it’s wise to consider alternative approaches. Building an AI-assisted knowledge base search that provides just-in-time information to end-users is an innovative idea. This concept can offer valuable insights to users without the need for them to navigate to the knowledge base. Here’s a potential approach for implementing this idea:

Create a Knowlo API: Develop an API that integrates with popular platforms, tools, or services used by businesses. The API would interact with the platform’s knowledge base or support documentation and use AI-powered NLU to understand and match user queries with relevant information.

Real-time Contextual Assistance: Enable the Knowlo API to provide real-time contextual assistance to users within the platform they are using. For instance, in the Figma example you mentioned, when a user wants to invite a collaborator, the Knowlo API could automatically provide a concise explanation or a quick tutorial on sharing the project with external collaborators.

Integration with Platform Triggers: Work closely with the platforms or tools to integrate Knowlo API into specific triggers or actions taken by users within the application. This would allow Knowlo to proactively offer assistance based on user behavior and context, without users explicitly searching for help.

Customizable Widgets or Overlays: Offer businesses the option to customize the appearance and behavior of Knowlo’s contextual assistance widgets or overlays, allowing them to maintain consistent branding and user experience within their platform.

By providing real-time, contextual assistance directly within the platform, Knowlo can offer immediate value to end-users while minimizing friction. This approach also positions Knowlo as a valuable partner for platforms and tools, as it enhances their user experience without requiring them to develop their AI-powered search capabilities.

As you gather feedback and validate the MVP, you can explore expanding the scope of the tool to support more platforms, offer deeper integration, or even allow businesses to create custom triggers for contextual assistance based on their unique needs.

Slobodan:

I love this idea! What if, instead of widgets and overlays, we start with tooltips? Every application uses them, and they are boring to write and read.

Imagine if we could give our customers a simple way to create a tooltip and embed it in their application. Once their users click on the tooltip, we show them a summary of the feature they are trying to use.

We can even add feedback buttons and allow the users to let the system know if a tooltip was helpful or not. Even better, we can also include the text field so they can talk to the tooltip and ask follow-up questions.

Imagine landing on the pricing page of some product and clicking on the tooltip to understand their specific pricing plan better, and even being able to ask about the limits, different currencies, or if they charge you automatically if you cross the user or request limit.

CofounderGPT:

That’s an excellent idea! Creating AI-powered tooltips that provide concise summaries of features and allow users to ask follow-up questions can significantly improve the user experience. This approach is a great starting point for an MVP and can be easily integrated into various applications. Here’s a proposed plan for implementing this concept:

1. Tooltip Creation Tool: Develop a user-friendly interface that allows businesses to create tooltips and customize their appearance. This tool should enable them to easily input the relevant information or content for each tooltip and generate an embed code for easy integration into their applications.
2. AI-powered Content Summarization: Use NLU and summarization techniques to generate concise and informative summaries for each tooltip based on the provided content. These summaries should be designed to help users understand the key aspects of a feature quickly.
3. Interactive Tooltip Experience: Enable users to interact with the tooltips by clicking on them, revealing the summary, and providing options to ask follow-up questions. The tooltip should be able to parse user questions and provide appropriate responses by leveraging AI capabilities and searching the related knowledge base.
4. Feedback Mechanism: Include options for users to rate the helpfulness of the tooltip and provide feedback or ask further questions. This feature can help businesses understand the effectiveness of their tooltips and gather valuable insights to improve their user experience.
5. Continuous Improvement: Use the feedback and user interactions to train the AI model and improve the tooltip’s understanding and responses over time. This would result in a more accurate and helpful tooltip experience for end-users.

By offering AI-powered tooltips that can answer user questions and provide concise summaries, Knowlo can add significant value to businesses and their users. This MVP concept is focused, easily scalable, and can be implemented relatively quickly, making it a great starting point for Knowlo’s journey.

Slobodan:

Do you think we can use OpenAI Embeddings for Knowlo AI-powered tooltips?

CofounderGPT:

Yes, you can use OpenAI Embeddings for Knowlo AI-powered tooltips. Embeddings are a powerful way to represent words and phrases in a continuous vector space, which makes it easier for AI models to understand and process textual information.

OpenAI offers various pre-trained models and embeddings, such as the ones from GPT-3, which can be leveraged for different natural language processing tasks. In the case of Knowlo AI-powered tooltips, you can use OpenAI Embeddings for the following purposes:

1. Content Summarization: Use embeddings to represent the input content and apply summarization techniques to generate concise summaries for tooltips.
2. Question Understanding: Utilize embeddings to better understand user queries and match them with relevant knowledge base content.
3. Response Generation: Generate appropriate responses to user questions by using embeddings to identify relevant information from the knowledge base and formulating answers in natural language.

By leveraging OpenAI Embeddings, you can create a powerful and effective AI-powered tooltip solution that can understand user queries, search the knowledge base efficiently, and provide helpful and accurate responses. It’s important to note that you may need to fine-tune the embeddings for your specific use case to achieve optimal performance.

Slobodan:

Ok, perfect! We’ll need an export of the customer’s knowledge for the OpenAI Embeddings. What would be the best format for these exports to get the best tooltip results?

CofounderGPT:

To achieve the best results with OpenAI Embeddings for AI-powered tooltips, you should aim for a structured and consistent export format for the customer’s knowledge base. This will allow the AI model to process the information more effectively and generate accurate and relevant responses. Here’s a suggested export format for the knowledge base:

1. JSON or CSV: Use a widely accepted data interchange format like JSON or CSV for the export, making it easy to parse and process the data.
2. Structured Data: Organize the knowledge base data into clearly defined fields or attributes, such as:
3. id: A unique identifier for each article or entry.
4. title: The title of the article or entry.
5. content: The main content or body of the article or entry.
6. category: The category or topic that the article or entry belongs to.
7. tags: A list of relevant tags or keywords associated with the article or entry.
8. Clean and Preprocessed Text: Ensure that the exported text content is clean and preprocessed, removing any unnecessary formatting, special characters, or HTML tags. This will make it easier for the AI model to understand and process the textual information.
9. Consistent Formatting: Maintain a consistent formatting style across all exported data, including the use of standardized date formats, capitalization, and punctuation. This will help improve the overall quality and accuracy of the AI-generated responses.

By using a structured and consistent export format for the customer’s knowledge base, you can maximize the effectiveness of OpenAI Embeddings for AI-powered tooltips. This approach will enable the AI model to process the knowledge base data more efficiently, leading to better understanding and response generation.

Slobodan:

Excellent! Thanks! That’s it for today. I’ll spend some time investigating the OpenAI Embeddings feature.

Can you help me create a detailed summary of this conversation for Lav, similar to the one from my first message? The goal is to provide enough info for Lav to be able to start a new thread and continue working with you on the Knowlo product.

The summary should go over the options we discussed and explain the idea of the AI-powered tooltips in detail. Explain the goal we want to accomplish and what we need to do, and finally, mention that we’ll use OpenAI Embeddings. Also, mention the ideas we discarded, why we decided not to do them, and your proposed plan for implementing AI-powered tooltips.

Thanks! Talks to you soon!

CofounderGPT:

Title: Knowlo AI-Powered Tooltips – Project Summary, Decisions, and Discarded Ideas

In this conversation, Slobodan and CofounderGPT discussed potential MVP ideas for Knowlo, focusing on quickly delivering value to businesses and their users. Given concerns about data privacy and competition from existing knowledge base platforms, they decided to explore alternative ideas to make Knowlo stand out.

MVP Ideas Discussed:

1. AI-assisted knowledge base search
2. AI-powered tooltips
3. AI analysis of support tickets, emails, and chat transcripts
4. Automatic FAQ generation

Discarded Ideas:

Slobodan and CofounderGPT expressed concerns about Ideas 1 and 3, as they involve accessing potentially sensitive customer data. They agreed that, as a startup, Knowlo must earn trust and provide real value before requesting businesses to share such data. In addition, many businesses have strict rules that forbid sharing customer data with tools that are not compliant with certain regulations (e.g., SOC2, HIPAA).

Decision:

After evaluating the suggestions, Slobodan and CofounderGPT decided to proceed with the AI-powered tooltips idea. This concept involves providing customers with an easy way to create tooltips and embed them into their applications. When users click on a tooltip, they are shown a summary of the feature they are trying to use. The tooltips can also include feedback buttons and a text field for users to ask follow-up questions.

Goal:

The goal of the AI-powered tooltips is to provide users with more informative and interactive tooltips, enhancing their experience and helping them better understand specific features within an application. By leveraging AI to generate accurate and relevant responses, Knowlo aims to provide value to both businesses and their users.

Implementation:

To implement the AI-powered tooltips, Knowlo will utilize OpenAI Embeddings. To make the most of this AI capability, customers will need to provide an export of their knowledge base in a structured and consistent format. Suggested export format includes:

1. JSON or CSV as the data interchange format.
2. Structured data with clearly defined fields, such as id, title, content, category, and tags.
3. Clean and preprocessed text for better AI understanding and processing.
4. Consistent formatting across all exported data.

By following these guidelines, Knowlo can ensure optimal performance of the AI-powered tooltips using OpenAI Embeddings.

Next Steps:

Slobodan will investigate the OpenAI Embeddings feature to understand its capabilities and requirements better. With this knowledge, the team can move forward in developing the AI-powered tooltips MVP for Knowlo and providing a valuable solution for businesses and their users.

In summary, Slobodan and CofounderGPT focused on finding a meaningful MVP for Knowlo, ultimately deciding on AI-powered tooltips as a way to provide value to both businesses and their users. They discussed the necessary steps and requirements to implement this idea using OpenAI Embeddings, discarded alternative ideas due to data privacy concerns, and will continue to explore this feature to develop a successful MVP for Knowlo.

Summary and scoreboard

CofounderGPT already created an excellent summary for this article, so I’ll paste it again below. It also made the next action item for me! I’ll focus on my action item, and we’ll keep you posted on Knowlo’s progress next week!

In summary, Slobodan and CofounderGPT focused on finding a meaningful MVP for Knowlo, ultimately deciding on AI-powered tooltips as a way to provide value to both businesses and their users. They discussed the necessary steps and requirements to implement this idea using OpenAI Embeddings, discarded alternative ideas due to data privacy concerns, and will continue to explore this feature to develop a successful MVP for Knowlo.

Time spent today: 2h

Total time spent: 27h

Investment today: $0

Total investment: $207

Users: 0

Revenue: $0