What is Cognita
Cognita is an open-source modular RAG (Retrieval Augmented Generation) framework that helps developers build modular, easily extensible, and deployable production-grade applications. Cognita uses Langchain/LlamaIndex as its underlying technology and provides a code organization structure, allowing each component to be API-driven, easy to manage, and extensible. Cognita supports local deployment, provides a production-ready environment and no-code UI support, and supports incremental indexing by default. Based on Cognita, users can easily test and deploy RAG systems, and support non-technical users in interacting with the system through the UI.
Main features of Cognita
- Modular and API-driven architecture: Each component of Cognita is modular and can interact through APIs, making integration and extension easy.
- Local and production environment support: Supports quick setup and testing in local environments and provides production-level deployment options.
- No-code UI: Provides a user interface that supports non-technical users in uploading documents and performing question answering without writing code.
- Incremental indexing: Supports incremental indexing, efficiently handles document updates, and reduces computational burden.
- Multi-document retriever support: Supports multiple document retrieval techniques, such as similarity search, query decomposition, and document reranking.
Technical principles of Cognita
- Data indexing: Based on indexing jobs, periodically scans files in data sources, parses them into small chunks, generates embedding vectors with embedding models, and stores them in a vector database.
- Query service: Uses an API server (such as FastAPI) to handle queries and generate answers. This involves using an AI gateway to process user queries, interacting with the vector database to retrieve relevant information, and ultimately using an LLM to generate answers.
- LLM/embedding model deployment: Deploys LLMs and embedding models as separate services and calls them based on APIs.
- Vector database deployment: Supports deploying vector databases in production environments, supporting large-scale and reliable data storage and retrieval.
- RAG components: Decomposes the RAG system into multiple components, including data sources, metadata storage, LLM gateway, vector database, indexing jobs, and API server. Each component can be configured and managed through the UI or API.
Cognita project address
- GitHub repository：https://github.com/truefoundry/cognita
Application scenarios of Cognita
- Question answering systems: Build enterprise-level question answering systems that support users in querying specific information and obtaining accurate answers from large numbers of documents.
- Knowledge management: Help organizations manage and retrieve internal knowledge bases, including policy documents, operation manuals, and FAQs.
- Customer support: In customer service scenarios, automatically answer common questions to improve efficiency and reduce response time.
- Content recommendation: Recommend related content, such as articles, videos, or products, based on user queries and historical behavior.
- Research and analysis: Provide tools for researchers to help users quickly find relevant information from large numbers of research documents.