<aside>
💡 This is a WIP documentation, for any feedback write to [email protected]
</aside>
Tech Description
Brian Knows is composed of two main components that work together to provide users with a powerful and comprehensive knowledge base:
- NucliaDB: this tool serves as the foundation of Brian Knows. It is a powerful vector database designed specifically for aggregating and transforming extensive documentation, articles, and online resources related to projects and protocols in the web3 space. NucliaDB acts as a centralized repository, organizing and storing information in a structured and easily accessible manner.
- GPT-3 Model: Brian Knows utilizes the state-of-the-art GPT-3 (Generative Pre-trained Transformer 3) language model developed by OpenAI. GPT-3 is trained on the vast amount of information available inside the NucliaDB vector database. It enables users to perform searches and retrieve relevant information by understanding natural language queries and generating accurate responses.
How Information is Analyzed
The primary objective of Brian Knows is to consistently update and provide users with the most pertinent and current information related to web3. The process of analyzing and delivering this information involves several key steps:
- Data Acquisition: Brian Knows acquires diverse forms of unstructured data from a wide range of sources. This data includes documentation, articles, whitepapers, blog posts, forums, and other relevant web resources. These sources are carefully selected to ensure a comprehensive coverage of the web3 ecosystem.
- Data Ingestion and Normalization: The acquired data is ingested into the NucliaDB vector database. During the ingestion process, the data undergoes normalization, ensuring consistency and standardization across different sources. This step involves transforming the data into a common format and resolving any discrepancies or inconsistencies.
- Content Extraction and Indexing: Once the data is loaded into NucliaDB, the system extracts, indexes, and organizes the content. This includes identifying entities, tokens, paragraphs, classifications, previews, and metadata associated with each piece of information. This comprehensive indexing allows for efficient and accurate retrieval of relevant data during user queries.
- Natural Language Understanding: Brian Knows leverages the power of the GPT-3 model to understand and interpret user queries expressed in natural language. GPT-3 has been trained on a diverse range of text data, enabling it to comprehend the meaning, context, and intent behind user requests. This natural language understanding capability enables users to interact with Brian Knows in a conversational manner, just as they would with a human expert.
- Response Generation: Once the user query is understood, Brian Knows generates accurate and informative responses. The system utilizes its vast knowledge base stored in NucliaDB to provide relevant information, insights, and recommendations. The response generation process takes into account the context of the user query and provides clear and concise answers, assisting users in their exploration of the web3 landscape.
Continuous Improvement
Brian Knows is an evolving system that continuously expands its knowledge and improves its capabilities. The development team actively engages in ongoing research and development efforts to enhance the performance, accuracy, and coverage of the system. This includes regular updates to the NucliaDB database with the latest information and advancements in the web3 space, as well as refining the GPT-3 model through fine-tuning and incorporating user feedback.