Beginner8 min read2 of 40

Simple RAG

The foundation — load a PDF, split into chunks, embed with Bedrock Titan, store in Qdrant, and retrieve by similarity.

Overview

Property	Detail
Pattern	Simple RAG — The Foundation
Level	1 (Entry point)
Embeddings	AWS Bedrock — Amazon Titan Embed Text v2 (1024 dims)
LLM	AWS Bedrock — Claude Sonnet (cross-region inference)
Vector DB	Qdrant Cloud (eu-west-2, cosine similarity)
Framework	LangChain
Notebook	01_Simple_RAG.ipynb

Architecture

flowchart TD
    subgraph INGEST["📥 Indexing Pipeline"]
        A[PDF Document] --> B[Split every 1000 chars]
        B --> C[Embed with Bedrock Titan]
        C --> D[(Qdrant Cloud)]
    end

    subgraph QUERY["🔍 Query Pipeline"]
        E[User Query] --> F[Embed query with same model]
        F --> G{Cosine Similarity Search}
        D --> G
        G --> H[Return top-2 most similar chunks]
    end

How It Works

Step	Action	Detail
1	Load PDF	PyPDF extracts text from document
2	Chunk	Split every 1000 characters (fixed size)
3	Embed	Amazon Titan Embed Text v2 → 1024-dim vectors
4	Store	Upload vectors to Qdrant Cloud
5	Query	Embed question → cosine similarity search → top-2 chunks

Test Output

Query: "What causes climate change?"

→ Returns the 2 most similar chunks from the PDF
  that discuss causes of climate change.

Think of it as: A search engine for your documents.

Tech Stack

Component	Technology
Embeddings	AWS Bedrock — Amazon Titan Embed Text v2 (1024 dims)
LLM	AWS Bedrock — Claude Sonnet
Vector DB	Qdrant Cloud (cosine similarity)
PDF Reader	PyPDF
Framework	LangChain
Config	Centralized `.env` file

Limitations

Limitation	Impact	Solution
Fixed 1000-char splits	May cut mid-sentence	→ Semantic Chunking
No relevance check	Returns whatever is "similar"	→ Self-RAG
No quality filter	Top-K may include noise	→ Reranking
No verification	LLM may hallucinate	→ Reliable RAG

When to Use

Scenario	Use Simple RAG?
Quick prototype / POC	Yes
Learning RAG concepts	Yes
Production with high accuracy needs	No — add reranking + verification
Long documents with topic shifts	No — use semantic chunking

Source

▸01_Simple_RAG.ipynb

Related Concepts

Semantic Chunking

Same as Simple RAG but with smarter scissors — splits where the topic actually changes instead of every 1000 characters.

Reranking

The quality filter — retrieves 10 docs, then a second model re-scores them. Two methods: fast cross-encoder or smarter LLM reranking.

Self-RAG

The smart decision maker — decides if retrieval is needed, filters irrelevant docs, generates multiple answers, and picks the best one.