Retrieval laboratory · zero dependencies
See exactly what your RAG pipeline retrieves
Configure the same character chunking, overlap, hash embedding, cosine similarity search, and prompt assembly exposed by RAGPipeline. Every result is traceable to its source document and chunk index.
DocumentLoader→TextChunker→embed_text→VectorStore.search→PromptAssembler
Bundled corpus
Four realistic product and operations documents.
Ask the corpus
Try deployment, authentication, backups, monitoring, or architecture.
Retrieval results
Ranked by cosine similarity against the query vector.
Assembled prompt
Output of PromptAssembler.assemble().
Generated answer
Grounded only in retrieved context.
🔒 This demo runs on a bundled 4-document corpus. The full pipeline ingests your real docs at any scale.
src/rag_pipeline.py— Single-file RAG pipeline: document loader, text chunker, vector store, retrieval engine, prompt assembler- Document loader — Ingest .txt, .md, .py, .json, .csv files from a directory
- Text chunker — Configurable chunk size (200-2000) and overlap (0-500)
- Vector store — In-memory with cosine similarity search, top-K retrieval
- Prompt assembler — Template-based with context injection, configurable system/user prompts
- CLI + API — Use from terminal or import as a library, built-in demo mode
- Zero dependencies — Python 3.10+ stdlib only