rag-from-scratch and SRAG
These are complements: the educational framework for understanding RAG fundamentals pairs naturally with an advanced production system that implements those concepts at scale with specialized capabilities like audio processing and sophisticated document parsing.
About rag-from-scratch
pguso/rag-from-scratch
Demystify RAG by building it from scratch. Local LLMs, no black boxes - real understanding of embeddings, vector search, retrieval, and context-augmented generation.
Implements a modular, JavaScript-based RAG pipeline with progressive learning examples covering embeddings, in-memory vector indexing, and retrieval strategies including hybrid search, multi-query decomposition, and query rewriting with LLM fallbacks. Built entirely with local models (via node-llama-cpp) and includes reusable library components for caching, normalization, and result fusion techniques like reciprocal rank fusion.
About SRAG
CyrilDesch/SRAG
Highly flexible RAG system with advanced document parsing and audio processing.
Implements hybrid retrieval combining vector search, BM25 lexical search, and cross-encoder reranking with metadata filtering. Built on Scala 3 and ZIO with hexagonal architecture, it decouples domain logic from pluggable adapters (PostgreSQL, Qdrant, OpenSearch, Whisper, MinIO, Redis) configurable entirely via environment variables for infrastructure-agnostic deployment. Designed for stateless horizontal scaling in production rather than single-machine experimentation, with built-in audio transcription and vision-based document extraction pipelines.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work