NearNDear | Devpost

Demo

Inspiration

High Entry Barriers to Creating Custom LLMs

Many of us have experienced going back and forth with an LLM to get the answer we want. Ideally, we’d fine-tune the LLM to respond precisely to our needs, but this requires significant prior knowledge, like coding and AI expertise. Near N Dear aims to make this process easy and accessible, bringing customization close to us.

The Ownership Dilemma in Custom LLMs

Let’s take a look at apps that let us create Custom GPTs (like a GPT Store). Here, we can build custom LLMs relatively easily. But the problem is, even after customizing it, we have no control over it. We can’t see how often our GPT is used, how much revenue it generates, or gain any benefit when others use our creation. All the profits go to OpenAI. Near N Dear aims to return full ownership of Custom LLMs to creators, allowing them to earn revenue and regain rightful control over their creations.

Uncertain Sources Yield Unreliable Responses

The biggest drawback of LLMs is hallucination—they sometimes provide answers that aren’t based on facts, requiring users to double-check responses for accuracy. Near N Dear aims to make LLM responses trustworthy by clearly indicating sources, allowing users to have confidence in the answers provided.

What it does

Breaking Down Barriers: Making Custom LLMs Accessible to All Without Coding

With Near N Dear, customizing an LLM is entirely code-free. No need for coding skills or advanced AI knowledge—just type in your instructions, and we’ll take care of the rest. Unlike other platforms that charge subscription fees for customization, Near N Dear offers it all in a single, one-time transaction, making it much more affordable and accessible.

Improving Accuracy & Minimizing Bias

Instead of depending on general LLMs, our platform uses RAG to allow each AI to access a specialized knowledge base that is curated by its creator. RAG dynamically retrieves relevant information from the creator's custom datasets, augmenting the generation process with specific, contextually accurate insights.

Thus, creators can build tailored AIs in their areas of expertise, which use RAG to pull precise and relevant information from their expert sources, ensuring that responses are highly accurate and contextually deep. Rather than relying on generalized models, each interaction taps into specialized data repositories, dynamically retrieving the most relevant information curated by experts in the field.

Empowering Control and Transparency for Creators

Our platform redefines this approach by ensuring creators retain full ownership of their data and creations, with complete transparency at every stage. Powered by the NEAR blockchain, we give creators visibility into data usage and decision-making, guaranteeing their work is used only with explicit consent and fair compensation. This commitment to control and transparency fosters trust, accountability, and a collaborative AI ecosystem that truly values its creators.

Ensuring Fair Compensation for Every Contribution

Our platform is designed to ensure that all contributors—whether providing data, content, or feedback—are fairly rewarded. Through a token-based revenue-sharing model, we’re building a more balanced ecosystem that aligns the interests of creators, users, and the platform alike.

How we built it

RAG(Retrieval-Augmented Generation) based Custom LLM

When a creator inputs the content they want to customize and initiates a transaction, we handle data chunking and preprocessing based on the category, generate embeddings, and store them in a Vector DB. Then, when a user asks a question, we retrieve the most similar embedding to provide a relevant answer. All the user has to do is input the text, and we take care of setting it up for the best responses.

Transparency Enhanced with NEAR Blockchain

We transparently publish the data used by creators for customization on the blockchain. This allows users to see exactly what data went into creating each custom model, enhancing trust in the answers and giving insight into which data yields the best responses. While AI often operates as a black box, using Near N Dear in this way can help users gradually build a better understanding of AI.

Reward Systems Based on NEAR Tokenomics

When a user utilizes a customized LLM, a portion of the usage fee goes to creators who have put in their time and effort to customize it. With our seamless and sophisticated token-based NEAR revenue policy, our platform strongly incentivizes creators to make quality customizations for better earnings, while users instantly monetize by simply utilizing our well-customized models without the need of any additional tuning.

Challenges we ran into

Integrating Blockchain & AI

As we set out to develop Near N Dear, we quickly realized that the lack of established decentralized custom AI platforms in the real world added a significant layer of difficulty to our project. While numerous AI platforms exist, few—if any—successfully combine AI capabilities with the decentralization principles of blockchain.

This gap in the market meant that we had limited reference points or case studies to guide our integration efforts. Most existing AI solutions are built on centralized architectures, making it challenging to adapt their functionalities to a decentralized framework. We had to innovate from the ground up, figuring out how to maintain the advantages of AI, such as advanced data processing and learning capabilities, while also adhering to the principles of decentralization and data ownership inherent to blockchain technology.

Speeding Up Response Times

In developing Near N Dear, we recognized that speed is a critical factor for both AI and blockchain applications. To provide a user experience that rivals that of traditional Web2 applications, fast processing and response times are essential. However, integrating AI with blockchain posed significant challenges in achieving the rapid performance we aimed for.

AI algorithms, particularly those that require extensive computations, often demand substantial processing power and time to deliver accurate responses. Meanwhile, blockchain transactions, while secure and transparent, inherently involve a layer of latency due to the consensus mechanisms required to validate and confirm each transaction. The combination of these two technologies created a bottleneck, where the advantages of decentralization and security offered by blockchain could slow down the rapid data processing expected from AI.

WEB3 Chat UX

In conventional messaging platforms like Telegram or Instagram, users can send messages seamlessly without generating a transaction for each interaction. This frictionless experience is essential for user satisfaction and engagement, as it allows for spontaneous and fluid communication.

However, in a Web3 context, every interaction often necessitates a transaction on the blockchain. This requirement, while crucial for enabling features like rewards and ensuring data integrity, introduces a layer of complexity that can hinder user experience. Users may find it cumbersome to deal with transaction confirmations and gas fees every time they send a message or engage in a conversation, which can detract from the immediate and effortless nature of communication that they expect from traditional platforms.

Accomplishments that we're proud of

Improved performance compared to GPTs

One of our significant accomplishments with Near N Dear has been the development of customized large language models (LLMs) that consistently outperform standard ChatGPT, particularly in areas requiring the latest information or addressing niche topics with limited online resources.

Through our innovative approach, we empower creators to tailor their LLMs, allowing them to infuse their unique expertise and specialized knowledge into the model. This customization enhances the model's relevance and accuracy, enabling it to provide more insightful and contextually rich responses than generic versions of GPT. This performance improvement can be attributed to our use of advanced techniques such as Retrieval-Augmented Generation (RAG), which allows our LLMs to access curated datasets specific to their domains. By dynamically retrieving relevant information, the models can deliver highly accurate and context-aware responses that better serve user needs.

Our Seamless Chat UX

We recognized that for our platform to attract and retain users, it was essential to deliver a chat interface that feels as intuitive and immediate as popular messaging apps. We aimed to replicate the effortless interaction users enjoy in platforms like ChatGPT, where responses are instantaneous and engaging, without the added friction of blockchain transactions. This meant that we needed to overcome the inherent complexities of Web3 technology, particularly the need for transaction confirmations that can interrupt the flow of conversation.

To achieve this, our team focused on developing a system where users can engage in conversations without needing to approve a transaction each time they send a message. By streamlining the transaction process, we ensured that users could communicate fluidly, receiving responses in real time—just like they would in a typical chat environment. This accomplishment required innovative thinking and technical solutions, such as implementing batch transactions and optimizing backend processes, allowing us to abstract the complexities of blockchain from the user experience.

What we learned

The Synergy of Blockchain and AI

Through our development journey, we discovered that the integration of blockchain technology can effectively address several critical challenges faced by artificial intelligence. One of the most pressing issues in AI is the black-box problem, where the decision-making processes of models remain opaque to users. This lack of transparency often leads to mistrust and hesitancy among users. By leveraging blockchain, we can provide an immutable record of data inputs and decision pathways, allowing users to trace how outcomes are generated. This transparency fosters trust and accountability, empowering users to understand and validate the AI's decisions.

The synergy of AI and blockchain also paves the way for enhanced ownership of data and intellectual property. In conventional AI platforms, users frequently sacrifice their control over the data they provide, leading to a disconnect between creators and their creations. With blockchain's robust architecture, we can empower creators to retain full ownership of their data and AI models. This shift not only allows creators to monetize their contributions fairly but also ensures that their work is used only with explicit consent.

Outstanding NEAR blockchain compared to other blockchain

As we developed Near N Dear, we came to appreciate the unique strengths of the NEAR blockchain, which proved essential to our success. What sets NEAR apart from other blockchain protocols is its exceptional transaction speed and scalability, allowing us to execute complex operations swiftly without compromising performance. This capability was particularly beneficial for our AI-driven platform, where rapid data processing and real-time interactions are crucial.

Moreover, NEAR’s well-developed SDKs specifically tailored for AI applications provided us with invaluable tools and resources. The ease of use and extensive documentation offered by NEAR enabled us to build and deploy our platform efficiently, streamlining the development process and accelerating our time to market.

What's next for NearNDear

Advanced RAG Algorithm: Expanding Input Capabilities

At Near N Dear, we currently accept text as input data, leveraging our advanced Retrieval-Augmented Generation (RAG) algorithm to generate tailored responses. However, our vision extends far beyond text alone. In the future, we plan to enhance our platform to support a wider array of input formats, including images, PDFs, and graphs. This expansion will allow users to provide diverse data types, enriching the context and depth of the AI's understanding.

To facilitate this, we are committed to continuously refining our data chunking methods. Improved chunking will enable our algorithm to effectively segment and process complex information, ensuring that even large datasets or intricate visual elements can be interpreted accurately. By optimizing how data is retrieved and utilized, we aim to enhance the reliability and relevance of the responses generated by our AI models.

Dashboard for creators: Feeding Insights for Engagement

We are committed to empowering creators with full ownership and control over their custom LLMs, and a key component of this initiative is our enhanced dashboard. In the future, creators will have access to detailed analytics, including target demographics, peak usage times, and engagement metrics. This information will enable creators to better understand their audience and tailor their customizations accordingly, optimizing their AI models for maximum impact.

Additionally, we plan to incorporate features that allow creators to track performance trends over time, analyze user interactions, and gather feedback, all of which will contribute to refining their models. By providing these insights, our dashboard will not only facilitate better decision-making but also foster a sense of community and collaboration among creators, as they can learn from one another and share best practices.

Cultivating a Creator-Driven AI Community

We believe that decentralization fosters greater innovation by enabling collaboration across a global community. Unlike conventional platforms, where innovation is stifled by corporate control, our approach encourages contributions from a wider range of content creators and users. This creates an open and dynamic ecosystem where new ideas can flourish, ultimately leading to a platform that evolves and improves based on the collective efforts of its AI creator community.