Skip to main content

Architecture

The system is built around five core components:

1. Parsers​

Convert various data formats into a universal Document format.

  • CSVParser: Generic CSV parser with configurable fields
  • CustomerSupportCSVParser: Specialized for support ticket data
  • PDFParser: Extract text, metadata, and structure from PDFs

2. Embedders​

Generate vector embeddings from text content.

  • OllamaEmbedder: Uses local Ollama models

3. Vector stores​

Store and search document embeddings.

  • ChromaStore: ChromaDB integration with persistence

4. Universal retrieval strategies​

Database-agnostic strategies that optimize for your store.

  • BasicSimilarityStrategy
  • MetadataFilteredStrategy
  • MultiQueryStrategy
  • RerankedStrategy
  • HybridUniversalStrategy

5. Pipeline​

Chains components together for end-to-end processing.