Production RAG recipe: semantic (embedding-based) chunking, cohere-embed-v3 embeddings, LanceDB storage, Cohere Rerank 3.5 reranking. Includes retrieval evals.
Implement step-back prompting to improve retrieval recall for PDFs with tables using text-embedding-3-large + parent-document retriever.
Implement step-back prompting to improve retrieval recall for HTML docs using jina-embeddings-v3 + SPLADE sparse.
Implement step-back prompting to improve retrieval recall for code repositories using cohere-embed-multilingual-v3 + dense (cosine similarity).
Implement step-back prompting to improve retrieval recall for Notion pages using text-embedding-3-small + parent-document retriever.
Implement step-back prompting to improve retrieval recall for earnings call transcripts using gte-large-en-v1.5 + SPLADE sparse.
Implement step-back prompting to improve retrieval recall for legal contracts using nomic-embed-text-v1.5 + dense (cosine similarity).
Implement step-back prompting to improve retrieval recall for research papers using voyage-3-large + parent-document retriever.
Implement step-back prompting to improve retrieval recall for API reference docs using e5-mistral-7b-instruct + SPLADE sparse.
Implement step-back prompting to improve retrieval recall for customer interview transcripts using bge-large-en-v1.5 + dense (cosine similarity).
Implement step-back prompting to improve retrieval recall for financial filings (10-K/10-Q) using voyage-3 + parent-document retriever.
Implement step-back prompting to improve retrieval recall for regulatory filings using arctic-embed-l + SPLADE sparse.