AI RAG & Chroma
Document chunking, vector embeddings, and retrieval-augmented generation
✂️ Text Chunking ⓘ
RecursiveCharacterTextSplitter splitting on paragraph, line, word, and character boundaries.
Break documents into overlapping chunks of configurable size and overlap to preserve semantic context across boundaries.
Quick pick:
📐 Semantic Embeddings ⓘ
384-dimensional dense sentence embeddings (all-MiniLM-L6-v2) with cosine similarity.
Encode two sentences into normalized vectors and compute their cosine similarity independent of exact keyword overlap.
Quick pick:
🔍 Vector RAG QA ⓘ
4-stage RAG pipeline: Split → Ephemeral Chroma Index → Top-k Retrieval → Grounded GPT-4o-mini Answer.
Ask a question against the reference paragraph; the pipeline retrieves top-3 matching chunks and answers strictly from retrieved context.
Quick pick questions: