LLM With Efficiency
Last updated
Last updated
LLMs, by definition, are text-based language models trained on existing data with billions of parameters. Given the current state of technology, they may not be pertinent for real-time information analysis, such as evaluating a new crypto project that emerged two weeks ago; they must be assisted or supplemented to leverage their intelligence effectively.
All existing stored data from all crypto projects that we retrieve on the Internet is computed in the Next Gem AI MongoDB database, sent and processed by the most powerful AI models available on the market today; here is a non-exhaustive list of them: GPT4, GEMINI AI, MISTRAL AI, GROK AI. We categorize projects using these models based on defined criteria and pre-set rules trained and reviewed both by the AI themselves and experts in the crypto investing industry. Each model evaluates the overall reliability of a project and assigns a score to each one, utilizing a comprehensive data set. They are continually updated with all available information we can gather from the project’s website, documentation, papers, and more.
Our LLM back-end processes are developed in C# language or Node.JS (Typescript). We interact with the APIs of each AI company, sending them the extra data extracted from the scrapper (see section below): they are executed in a safe environment in cloud instances. Some extra instructions are provided to give them specific tasks, knowing what they have to do precisely: known as assistants, we use a relatively low LLM temperature to avoid creative outputs and ensure the models do as taught & instructed. The software process related to the AI's interaction waits for new projects to be sent to their execution state through internal HTTP requests for specific requests demands or select the fresh one on our MongoDB database directly by filtering projects not analyzed yet by each model we implement. To determine the final score of each project, here is a summary of the mathematical equation used to calculate it:
Where: • n is the total number of rules or factors (in this case 15: Team and Founders" Background, Technology, and Innovation, etc.).
• ωi is the weight assigned to the iᵗʰ factor, reflecting its importance in the overall analysis. The weights should be determined based on the relative importance of each factor to the success of crypto projects.
• fi is the score assigned to the iᵗʰ factor for a particular project, based on its performance or status in relation to that factor.
Each factor score fi is multiplied by its corresponding weight ωi, and then all these weighted scores are summed to produce the total project score. This approach allows for a nuanced and comprehensive assessment of each project, taking into account a wide array of critical success factors. The higher the Project Score, the more favorable the project is considered based on the analysis criteria.
