https://datatweets.com/courses/generative-ai/embeddings-and-semantic-search/Recent content in Embeddings & Semantic Search on DATATWEETSHugoenCopyright (c) 2025 DatatweetsSat, 27 Jun 2026 09:00:00 +0200https://datatweets.com/courses/generative-ai/embeddings-and-semantic-search/lesson-1-what-embeddings-are/Fri, 14 Nov 2025 09:00:00 +0200https://datatweets.com/courses/generative-ai/embeddings-and-semantic-search/lesson-1-what-embeddings-are/An embedding turns text into a list of numbers that represents its meaning. Build the intuition for the ‘meaning space’, see a real 384-dimensional vector, and watch related sentences score as similar even with no shared words.https://datatweets.com/courses/generative-ai/embeddings-and-semantic-search/lesson-2-generating-embeddings/Fri, 14 Nov 2025 09:00:00 +0200https://datatweets.com/courses/generative-ai/embeddings-and-semantic-search/lesson-2-generating-embeddings/Generate embeddings locally and for free with sentence-transformers. Load all-MiniLM-L6-v2, encode single strings and batches, inspect the 384-dimensional vectors, and learn when a hosted API like Voyage AI makes sense.https://datatweets.com/courses/generative-ai/embeddings-and-semantic-search/lesson-3-measuring-similarity-and-distance/Fri, 14 Nov 2025 09:00:00 +0200https://datatweets.com/courses/generative-ai/embeddings-and-semantic-search/lesson-3-measuring-similarity-and-distance/Turn two embeddings into a single similarity score. Learn cosine similarity, compute it with numpy and with sentence-transformers’ util.cos_sim, see why normalized vectors make it a dot product, and relate it to cosine and Euclidean distance.https://datatweets.com/courses/generative-ai/embeddings-and-semantic-search/lesson-4-guided-project-semantic-search/Fri, 14 Nov 2025 09:00:00 +0200https://datatweets.com/courses/generative-ai/embeddings-and-semantic-search/lesson-4-guided-project-semantic-search/Tie embeddings, batching, and cosine similarity together into a real semantic search engine over a support-FAQ dataset. Embed the corpus once, rank by similarity to any query, and wrap it in a reusable search function.