Source: http://www.poma-ai.com/docs/guides/

# Guides

Start here when you want the high-level story first. These pages explain the tradeoffs, frame the problem, and point you into the more structured learning and SDK sections.

## Featured guides

- [RAG chunking guide](/guides/rag-chunking/)
  Learn how common chunking strategies work, where they break down, and how POMA approaches chunksets differently.
- [Document ingestion guide](/guides/document-ingestion/)
  Walk through ingestion patterns, why RAG needs chunking after ingestion, and how common tooling stacks compare.
- [RAG architecture guide](/guides/rag-architecture/)
  How to design a production RAG pipeline end-to-end — ingestion, chunking, embedding, retrieval, and generation.
- [Context engineering](/guides/context-engineering/)
  Why context quality is decided at the document, not the prompt, and what the four layers of the pipeline each contribute.
- [Ingestion for context engineering](/guides/ingestion-context-engineering/)
  Why agents can't skim, and how structure-aware ingestion, chunksets, and sandwich ordering follow from that.

## For the stack you already run

Tool-specific series — start from the parser or the vector database you already operate:

- **The optimal chunker, by parser** — chunk the raw output of [Mistral OCR](/optimal-chunker-mistral-ocr), [LlamaParse](/optimal-chunker-llamaparse), [Azure Document Intelligence](/optimal-chunker-azure-document-intelligence), [Unstructured.io](/optimal-chunker-unstructured), [Docling](/optimal-chunker-docling), [Marker](/optimal-chunker-marker), [AWS Textract](/optimal-chunker-textract), or self-hosted [PaddleOCR-VL](/optimal-chunker-paddleocr-vl) without re-running OCR.
- **Optimal chunks, by vector DB** — what to embed into [Qdrant](/optimal-chunks-qdrant), [Pinecone](/optimal-chunks-pinecone), [Weaviate](/optimal-chunks-weaviate), [pgvector](/optimal-chunks-pgvector), [Milvus](/optimal-chunks-milvus), [Chroma](/optimal-chunks-chroma), [Turbopuffer](/optimal-chunks-turbopuffer), [Vespa](/optimal-chunks-vespa), [Elasticsearch](/optimal-chunks-elasticsearch), [OpenSearch](/optimal-chunks-opensearch), [Redis](/optimal-chunks-redis), [MongoDB Atlas](/optimal-chunks-mongodb-atlas), or [LanceDB](/optimal-chunks-lancedb) for retrieval that keeps context.
- **[Pipeline recipes](/pipelines/)** — the full parser × database matrix, end to end.

## Continue into Learn

- [Chunking](/learn/chunking/)
- [Ingestion](/learn/ingestion/)