user@devops:~$ cat README.md
Semantic Search RAG
# Description
Complete semantic search and Retrieval Augmented Generation (RAG) system built with Sentence Transformers, FAISS, and FLAN-T5. The pipeline loads 500 real BBC News articles (June 2025), generates 384-dimensional embeddings with all-MiniLM-L6-v2, indexes vectors in FAISS (IndexFlatIP + IndexIVFFlat), and enables semantic search by cosine similarity. Includes a full RAG pipeline that retrieves the most relevant documents and uses them as context for FLAN-T5-small to generate precise answers. Compares TF-IDF vs semantic embeddings (0% overlap in Top-5), visualizes embeddings with PCA and t-SNE, and analyzes similarity distribution across categories.
# Key features
$ Semantic embeddings with Sentence-BERT (all-MiniLM-L6-v2, 384 dimensions)
$ FAISS indexing: exact search (FlatIP) and approximate (IVFFlat) with 1.9x speedup
$ Full RAG pipeline: retrieve (FAISS) → augment (context) → generate (FLAN-T5)
$ Functional semantic search: queries by meaning, not just exact words
$ TF-IDF vs Semantic Embeddings comparison with overlap visualization
$ 5 visualizations: PCA, t-SNE, category heatmap, score distribution, comparison
$ Real dataset: 500 BBC News articles with 70+ categories
# Gallery
# Technologies used
