400
3x
USA
LLM
About the client
James (Sevan) Bomar III is a spiritual expert and coach in lifestyle advancement, finance, and self-development. Renowned as a serial entrepreneur, James is dedicated to creating social platforms and AI-driven solutions. With 170K subscribers on YouTube, James is not only a thought leader but also an investor who specializes in developing products to address major problems faced by society.
Our collaboration with James began by supporting his Woocommerce shop, where he sells courses and a range of lifestyle products. As our partnership evolved, he entrusted us with the creation of a platform for generating images on spiritual topics.
And then James came up with another innovative idea: “What if EVNE could build an AI-powered chatbot using a custom Large Language Model (LLM) to help people get answers to any question related to my area of expertise?”. He needed a team that understood deeply his domain and could solve even intricate issues comprehensively instead of simply writing code – and our team was a perfect fit.
That’s how Sibyl AI, the world’s first AI spiritual guide, was born.
Keep reading to discover how EVNE Developers turned this idea into reality, from a business model to a final solution.
Challenge
After years of collecting books, videos, courses, and all sorts of data on spiritual topics, James was looking to create a vast database that could be easily accessible to users interested in getting answers to spiritual questions. He needed a tech team focused on long-term collaboration that understands the intricacies of engineering with extended release cycles and can facilitate a smooth transition from the MVP phase to a full-scale product without the need to hire in-house. So he reached out to EVNE, his long-term software development partner, to build an AI-powered chatbot that could answer any esoteric question.
Solution
Sibyl AI is the world’s first spiritual LLM with a mission to give people access to ancient wisdom and help them grow spiritually. Sibyl leverages artificial intelligence and machine learning to provide tailored answers to user inquiries, leveraging ancient teachings and spiritual insights that James Bomar spent years collecting.
To make the solution possible, we unified 5 datasets in one system, each with 160K pages of metaphysical content, and ensured a direct connection with GPT-4 Turbo for instant response generation. Each dataset has its function:
- Sibyl is a core LLM trained in the field of metaphysics
- Ennealogy is a dataset with metaphysical metrics for knowing everything about self
- Etymology is a dataset with terms, symbols, scribes, and spiritual chants decoded
- Superhuman is a holistic health model trained on data related to ancient healing traditions
- Sovereign is a dataset for studying sovereign law and conspiratorial dealings
Result
Here are the main features of Sibyl AI:
Choosing the most relevant dataset to answer a prompt
Each dataset is trained on a specific material to better answer user inquiries, whether it’s about numerology, health, or astrology. The primary challenge lay in seamlessly directing user queries to the dataset that would generate the most accurate and insightful responses.
To choose the most suitable dataset, we implemented a two-layer verification process. In the first layer, the system meticulously analyzes the user’s prompt by studying dataset summaries. This initial step helps determine which dataset aligns most effectively with the nature of the question. The second layer involves an in-depth examination of keywords in the user’s prompts, allowing precise matching with the appropriate dataset.
By adopting this approach, Sibyl can confidently respond to user queries most efficiently and accurately, delivering a tailored experience that aligns with the user’s specific spiritual focus.
Optimizing response generation to a mere 0,5 seconds with Pinecone
The next challenge was to accelerate the response generation. Initially, we employed a custom approach when the LLM model analyzed all available data, dividing it into paragraphs known as embeddings.
Seeking an even more efficient solution, we transitioned to Pinecone, a powerful cloud-based service designed for searching, ranking, and organizing intricate data structures.
With Pinecone, upon receiving a user request, the system sifts through the dataset based on prompt keywords. Analyzing a hundred relevant paragraphs, Pinecone identifies the most fitting information. These selected paragraphs are then seamlessly transformed into a cohesive answer through ChatGPT, resulting in a remarkable reduction in response time to a mere 0.5 seconds at most.
By integrating Pinecone into our workflow, we’ve not only significantly enhanced the speed of information retrieval, shortening it from 30 seconds to 0,3-0,5 seconds, but also streamlined the process of generating coherent responses to user queries.
Enhancing vectorization quality
Finally, we had to optimize the quality of vectorization, which is a process of selecting the most relevant information from the entire dataset. To achieve this, we researched different AI models. We meticulously analyzed their functionalities, using plugins to assess their responses to requests and the output.
The main limitation we faced was ChatGPT’s inability to process substantial amounts of text. To solve this, we refined our vectorization process and shortened the length of paragraphs to get responses as quickly as possible.
Fast iterations to secure the leading edge in the market
We had to swiftly make changes to the product as AI-powered solutions progress quickly and it was imperative to be among the first ones to adopt new features and attract users.
We managed to update Sibyl efficiently by leveraging third-party APIs and introducing new functionality. Leveraging this approach, we could add new features in just a day or two, always ensuring the solution remains relevant on the market.
Value Delivered
400+ users increase before launch
Even without a marketing campaign, the platform effortlessly gained over 400 new users in the initial 2 months at no customer acquisition cost, which proved market fit and validated the MVP concept. The scalable architecture maintains system agility without complexities or excessive costs.
AI-powered knowledge platform
We trained AI on multiple datasets and delivered the ultimate knowledge platform that unifies all the spiritual data collected throughout the years. The product constantly grows and continues to evolve its capabilities, improving user experience and providing insightful answers to all spiritual questions.
Custom dashboard to track success
To allow full visibility into product success, we created an environment to allow the client to track product conversions, expenses, and revenue through a custom dashboard. It ensures a seamless experience, enabling the client to make informed decisions and optimize business strategies.
EVNE Developers have the skills and experience required to create advanced healthcare software tailored to your needs. Hire us to build secure, scalable, and compliant healthcare solutions that streamline processes, improve patient outcomes, and leverage your data effectively.
