🥤 RAGLite is a Python toolkit for Retrieval-Augmented Generation (RAG) with DuckDB or PostgreSQL

AI-powered automation system that transforms natural language requirements into production-ready Kubernetes manifests through three specialized CrewAI agents. System is able to execute manifest creation and editing, knowledge retrieval using RAG with late chunking, web search, and static and runtime validation of the configuration files.

Sentence-aware embeddings using late chunking with transformers.

This demo measures late chunking against standard embedding on two corpora — one where it hurts and one where it helps — producing concrete retrieval numbers that engineers can use to make an informed choice.

Advanced High-Fidelity RAG pipeline featuring Agentic Hierarchy Parsing for structural correction and Context-Sensitive Late Chunking for preserving document context.

A small, focused library for splitting Markdown documents into semantically coherent chunks for retrieval-augmented generation.

Rigorous evaluation of contextual retrieval techniques on FinanceBench: comparing 5 embedders × 4 chunking strategies with bootstrapped confidence intervals on FinMTEB and FinanceBench.