A curated collection of RAG (Retrieval-Augmented Generation) projects covering foundational to advanced techniques, including prompt routing, agentic reasoning, and HyDe-style retrieval.
| Project Name | Description | Techniques Used | Link |
|---|---|---|---|
| Dynamic Prompt-Aware RAG | Routes user queries to the best prompt template based on intent (e.g., summarize, compare). | Prompt routing, LLM-based intent classification, LlamaIndex | View |
| Agentic Multi-Doc RAG | Agent breaks query into sub-questions and retrieves from relevant document-specific indices. | ReAct agent, query decomposition, per-doc vector indices, web fallback | View |
| HyDe RAG (Single + Multi) | Generates hypothetical answers to improve retrieval; supports single and multi-doc averaging. | Hypothetical generation, embedding reranking, dot product similarity | View |
| Corrective RAG | Enhances answer reliability by verifying and correcting initial RAG output. Involves an evaluator agent that scores document relevance, a query rewriter for web fallback, and a generator that produces final answers from refined knowledge. Adapts dynamically based on confidence scoring (CORRECT, INCORRECT, AMBIGUOUS). |
Evaluator agent (embedding + LLM), knowledge strip refinement, LLM query rewriting, web search fallback, generator agent | View |
| HyPe RAG | Improves retrieval by indexing hypothetical questions instead of chunk embeddings. Transforms query–document matching into query–question matching for better alignment with natural user queries. | Precomputed hypothetical prompts, dense embedding index (FAISS), prompt-to-prompt retrieval, LlamaIndex/FAISS integration | View |
| HyQe RAG | Enhances traditional RAG pipelines by generating hypothetical queries for each document chunk, embedding them, and using these embeddings to improve retrieval relevance | Cosine similarity, Re-Ranking, Query to Query matching | View |
| Self RAG | Dynamically decides whether to use retrieved information and how to best utilize it in generating responses, aiming to produce more accurate, relevant, and useful outputs. | Retrieval decision, context filtering, support & utility scoring, prompt-based critique | View |
| Fusion RAG | Leverages multiple query reformulations, document retrieval, and robust re-ranking to generate high-quality, contextually rich responses using LLMs | Re-Ranking(RRF score), Query Reformulation | View |
| RAPTOR RAG | Organizes document chunks into a hierarchical tree using recursive clustering and abstractive summarization | Gausian Mixture Model(GMM), semantic chunking | View |
| Knowledge Graph RAG | Builds knowledge graph over the data and gives context aware responses | Graph data structure | View |
Built with ❤️ for scalable, modular experimentation in the world of RAG.