18 RAG Techniques, Benchmarked

Retrieval-Augmented Generation is not one algorithm — it's a toolbox. This lab walks through 18 techniques, from Simple RAG to Adaptive RAG, each scored on the same hard query. Pick a technique to watch its pipeline animate and see the extra step it adds, or open the leaderboard to compare all 18 at once.

Adaptive & self-correcting

Adaptive RAG

standard new step decision
QueryClassifyDECIDEStrategyNEWRetrieveGenerateFactualAnalyticalOpinionContextual
Benchmark score+0.36 vs Simple RAG · rank #1/18
Simple RAG
0.50
Adaptive RAG
0.86
The problem

No single retrieval strategy is best for every kind of question.

The technique

First classify the query (factual, analytical, opinion, contextual), then dispatch a specialised retrieval strategy tuned for that type.

  • Matches the method to the question type.
  • The clear winner of the benchmark (0.86).
  • Flexibility beats any one-size-fits-all pipeline.

When to use: Production systems facing diverse query types.

Scores come from a single complex test query (judged 0–1), so treat them as a teaching signal rather than gospel — the lesson is that different queries need different strategies. Ready to build one? Try the runnable RAG and Agentic RAG projects, or explore the 14 Pillars of Agentic AI.