Cognita's architecture comprises several components including data sources, a metadata store, an LLM Gateway, a Vector DB, an indexing job, and an API server. It also allows for customization of data loaders, embedders, parsers, and VectorDB. The article also outlines how to deploy Cognita with Truefoundry, use the RAG UI, and contribute to its open-source development. Future developments for Cognita include support for other vector databases, scalar and binary quantization embeddings, RAG evaluation and visualization, and RAG optimized LLMs.
Key takeaways:
- Cognita is an open-source framework that provides an organized structure to your RAG codebase, making it easy to test locally and deploy in a production-ready environment.
- The framework offers a central reusable repository of parsers, loaders, embedders, and retrievers, and is fully API driven, allowing for easy integration with other systems.
- Cognita supports multiple document retrievers, SOTA OpenSource embeddings and reranking, LLMs using Ollama, and incremental indexing that ingests entire documents in batches.
- It also provides a UI that allows non-technical users to upload documents and perform QnA using modules built by the development team, and supports hosting multiple RAG systems using one app.