Technical Architecture
- Data ingestion: documents are imported, divided into meaningful sections (ingestion and chunking), semantically encoded (embedding) and stored in a vector database.
- Retrieval: a retriever finds the most relevant text passages in the vector database through vector similarity.
- Generation: the generator combines the original query and the retrieved documents and generates a fact-based answer.
- Feedback: a feedback loop checks the quality of the answers and continuously improves the retrieval and generation components.