About Ketz: Bringing Expertise Back to AI

Inspiration

The idea for Ketz grew from closely observing the rapidly growing field of AI and its big impact on digital content. We saw two major issues emerging:

Impact on Content Creators

Powerful AI models can generate answers, but they often do so without proper permission, payment, or clear credit to the original works. This process frequently cuts out the original sources, undermining the traditional ways creators earn money from their content. We recognized this as a serious challenge, potentially diminishing the value of human-created work.

Impact on Trust Among AI Users

At the same time, we noticed a significant lack of trust among AI users – especially researchers, students, and professionals. They were frustrated by AI making up facts (hallucinations), providing shallow answers on specific topics, and giving responses without traceable sources. Discussions on platforms like Reddit highlighted this deep frustration with the reliability of general AI.

Bridging the gap with ketz.

Ketz was designed to be the solution: to create a new, mutually beneficial partnership between human expertise and AI. Our goal is to let creators directly earn money from their knowledge while giving users an AI tool they can truly trust, based on verified sources they have chosen.

What it does

Ketz creates a new, mutually beneficial partnership between human expertise and AI, letting creators directly earn money from their knowledge while giving users an AI tool they can truly trust, based on verified sources they have chosen.
Our project is built on a secure system using Retrieval-Augmented Generation (RAG) that operates as a dual-sided marketplace:

• For Creators: Ketz provides features like control over how their content is indexed and flexible earning options. They upload their content to a private "Ketz Vault."
  
• For Users: Consumers benefit from a basic AI chat interface that provides answers with clear credit. When a subscribed user asks a question, Ketz searches for the most relevant content from the creators the user follows. These sourced pieces are then given to a core AI model along with the user's question, with a strict instruction: "Answer this question using only the provided context and cite your sources." The AI then generates the final answer, which Ketz presents with clear citations pointing back to the specific creators and their original works.

How we built it

When building Ketz, we realized the main challenge wasn't just technical, but also a fundamental question: how to use AI's power without diminishing the value of human knowledge and original work. Our solution focused on securely using a technique called Retrieval-Augmented Generation (RAG). We understood that content should be used to find and cite information, but critically, it should never be used to train or change the underlying AI model. This became key to how Ketz protects intellectual property and acts as a responsible AI platform.

Using Bolt.new to drive the process, we were able to create Ketz to:

1. Content Ingestion: Creators upload their content to a private "Ketz Vault."
   
2. Indexing: This content is divided into smaller pieces and converted into a digital format (called embeddings), then stored in a separate, secure database that acts as an index, not for training.
   
3. Query Processing: When a subscribed user asks a question, Ketz searches this index for the most relevant content from the creators the user follows.
   
4. Augmented Generation: These sourced pieces are then given to a core AI model along with the user's question, with a strict instruction: "Answer this question using only the provided context and cite your sources."
   
5. Attributed Response: The AI then generates the final answer, which Ketz presents with clear citations pointing back to the specific creators and their original works.

Challenges we ran into

We faced several significant challenges during development:

• Preventing AI from Making Things Up: A key technical challenge was ensuring the AI sticks strictly to the provided content and doesn't invent information. This required careful instruction of the AI and thoughtful system design. In our first iterations it wouldn't search the vector database, choosing instead to hallucinate it's own facts.

• Protecting Intellectual Property: Creating a truly secure system where creator content is used only for finding and citing information (RAG) and never for training the AI model demanded precise architectural choices and strong security. This directly addresses the common worry among creators about AI using their content without permission.

• Growing Both Sides of the Platform: Bringing in enough high-quality creators and engaged users at the same time presented the classic challenge of needing both groups to grow together. Our strategy involves offering attractive benefits for early users, like lower fees and special promotions, to help overcome this initial hurdle.

• Managing AI Usage Costs: Since Ketz uses external AI models (LLMs), a constant challenge is managing and optimizing the costs associated with how much those models are used, especially as the platform grows rapidly.

• Building Trust: Gaining the trust of both creators, who are cautious about AI misuse, and consumers, who need reliable information, is an ongoing effort. Our focus on providing verifiable answers with clear credit and strong intellectual property protection is key to overcoming this lack of trust.

Accomplishments that we're proud of

We are proud to have developed a system that fundamentally re-intermediates human expertise in the age of AI. Our key accomplishments include:

• Successfully designing and implementing a secure RAG architecture that ensures creator content is used ethically, solely for verifiable answers, and never for AI model training.
  
• Creating a dual-sided marketplace that aligns the financial incentives for creators with the need for trustworthy information for users.
  
• Addressing the critical pain points of AI hallucinations and untraceable sources for consumers.
  
• Providing a novel and predictable monetization pathway for content creators, empowering them with control over their intellectual property.

What we learned

We learned that the core challenge in AI is not just technical, but also philosophical: how to leverage AI's power without diminishing the value of human knowledge and original work. Our learning crystallized around the importance of secure Retrieval-Augmented Generation (RAG) and the critical distinction between using content for retrieval versus for model training. We also learned the immense value of a dual-sided marketplace approach, where the value for consumers increases with more high-quality creators, and vice-versa, creating a powerful "network effect" that drives growth.

What's next for Ketz: Monetize Knowledge, Get Trusted AI Answers

Next for Ketz, we plan to continue refining the platform based on early adopter feedback, expand our content creator base to enrich the knowledge available to users, and explore strategic partnerships with publishers and content aggregators to further scale our impact. Our focus remains on enhancing the trust and value for both creators and consumers in the evolving AI ecosystem.

We need to find creator partners to start uploading their content so that we can test the system.

Exploring Open Content for Demonstrations and Value Proposition

To effectively demonstrate Ketz's unique value proposition and allow potential users and creators to experience its benefits firsthand, we plan to conduct a test run using openly available, public domain, and open-access books. This strategy is inspired by the comprehensive analysis of suitable open non-fiction sources, particularly those with permissive Creative Commons licenses (like CC BY or CC0), which allow for commercial use and adaptation.

By integrating content from repositories such as the Directory of Open Access Books (DOAB) and OpenStax, Ketz can showcase its core functionalities: providing verifiable, attributed AI answers grounded in high-quality, peer-reviewed material. This test run will allow us to:

Iterate and Refine: Gain valuable insights into data ingestion, indexing, and retrieval performance with diverse content types, allowing for continuous platform refinement.

Demonstrate Value: Visually and interactively show how Ketz mitigates AI hallucinations and provides traceable sources, offering a tangible example of "trusted AI."

Showcase Monetization Potential: While these sources are open, the demonstration can illustrate the mechanism of a content-driven, attributed AI, providing a clear vision of what Ketz offers to creators who choose to monetize their original content.

This initiative will serve as a powerful proof-of-concept, enabling us to gather user feedback, improve the platform's capabilities, and clearly articulate the benefits of Ketz in a practical, accessible way.

List of Challanges and other Hackathon info

Our bolt.new project: latest (v4): https://bolt.new/~/sb1-9b7zevhn previous: v3: https://bolt.new/~/sb1-wyprugy1 v2: https://bolt.new/~/sb1-8l3xdpzk v1: https://bolt.new/~/sb1-cjzn1t3k

Deploy Challenge

netlify email: ricky@gonzalezstrategy.com

Startup Challenge

org slug: dyjuzikbgfyjgnzuminm (ketz) project: tejjnueulkrhctjclbkn (ketz)

Custom Domain Challenge

I used Entri to get Ketz.site - not sure how to document this here

Make More Money Challenge

I tried to use revenuecat but my usecase (with multiple, changing subscriptions) was too challenging to implement fully. I plan to continue trying though!

Project: Ketz user: rickygonzalez@gmail.com

Built With

• entri
• openai
• react
• revenuecat
• supabase