This project houses a Retrieval Augmented Generation (RAG) LLM application built for robust and context-aware text generation. It leverages the combined power of LangChain for orchestration, MLflow for tracking and experimentation, DVC for version control, and RAGAS for evaluation.
- LangChain: Streamlines the data pipeline for retrieval and generation tasks.
- Qdrant: Vector Database to store embeddings of documents.
- MLflow: Manages experiments, tracks ML pipelines, and logs metrics.
- DVC: Facilitates version control and reproducibility of datasets and code.
- RAGAS: Offers comprehensive evaluation metrics for RAG systems.
RAGAS empowers you to assess your RAG system's performance through various metrics. The ones used in this app are:
- Answer Semantic Similarity: Measures the meaning similarity between generated and ground-truth answers (0-1, higher is better).
- Answer Relevance: Evaluates how pertinent the answer is to the prompt (0-1, higher is better).
- Answer Correctness: Assesses the factual accuracy of the generated answer (0-1, higher is better).
- Harmfulness: Detects harmful language and information in the output.