Understanding Epis Base
What is a Knowledge Epis Base?
The core of Epis Base is the Knowledge Base. This is a unique and private entity for each user, meaning only you have access to it, and it is private by design.
Think of your Knowledge Base as a personal library. Within this library, you have multiple Knowledge Sources, which are like individual books. These Knowledge Sources can come in various formats and from different origins.
Types of Knowledge Sources
Your Knowledge Base can be built using a variety of sources, including:
- Documents: PDFs, DOCX files, PPTX presentations, and other unstructured formats.
- External Sources: Content from websites or other online platforms.
- Developer Contributions: Direct integrations, such as connecting a GitHub repository to include code and documentation.
Customizing Your Knowledge Base
You have complete control over how you structure your Knowledge Base. You decide which sources to include, how to organize them, and how they interact to provide the most relevant information for your needs.
From a Knowledge Source, you can create AI Chats tailored to your specific context or design quizzes. While AI Chats can utilize multiple sources, quizzes can only be created from a single Knowledge Source.
AI Integration in Epis Base
Epis Base leverages a diverse array of Large Language Models (LLMs) to deliver efficient and secure AI-powered services. Our approach emphasizes data privacy by prioritizing local models and selectively utilizing proprietary models when necessary.
Model Selection Strategy
To maintain optimal performance and data confidentiality, Epis Base employs a dynamic model selection strategy:
-
Local Models: Primarily, we utilize open-source models that process data internally, ensuring user information remains secure.
-
Proprietary Models: In scenarios involving ambiguous knowledge bases or specific user requirements, we may temporarily engage proprietary LLMs to enhance processing accuracy.
AI Model Routing
Epis Base dynamically assigns tasks to appropriate AI models based on the nature of the knowledge source and system resource availability:
-
Proprietary Models: When necessary, we integrate cloud-based AI models to enhance performance, including:
- OpenAI Models
- Google Gemini Models
- Anthropic AI Models
-
Open-Source Models: Whenever possible, Epis Base prioritizes local and open-source models, including:
- LLaVA: Utilized for visual tasks such as image and diagram processing.
- LLaMA: A general-purpose AI model for text-based tasks.
- Microsoft’s Phi Models: Optimized for document processing.
Embedding and Reprocessing
For text processing, Epis Base employs embedding model routing to enhance search accuracy. This involves:
-
Understanding the knowledge source’s content type and optimizing document structure accordingly.
-
Monitoring retrieval performance: If retrieval scores decline, our AI reevaluates embeddings and reprocesses documents to improve accuracy.
Enterprise Solutions
For enterprise clients, EpisBase offers dedicated models tailored to specific organizational needs. Explore our enterprise solutions here.