Loading...
Loading...
Pure vector search loses exact keywords. Pure keyword search loses meaning. You will build the hybrid retrieval stack that serious teams run in production: dense plus sparse, fused with RRF, reranked by a cross-encoder.
Message a mentor about fit, prerequisites, or where to start. Replies come on WhatsApp, usually within a day.
Engineers are learning here from
Build a document Q&A system that combines dense semantic search with sparse keyword matching, fuses results with Reciprocal Rank Fusion, and boosts precision with a cross-encoder reranker. All running locally with Qdrant.
Hybrid retrieval with dense vectors, sparse keywords, RRF fusion, and cross-encoder reranking.
What you'll ship
What you'll learn
Curriculum
Document processing
Parse PDFs with PyMuPDF4LLM and chunk them so retrieval has context to work with
Hybrid retrieval
Generate dense plus sparse vectors, query Qdrant with RRF fusion, and rerank with a cross-encoder
Who it's for
who built a RAG demo and saw it fail the moment users typed exact product names or codes
moving from toy cosine similarity to the hybrid retrieval stack used in production
who know BM25 inside out and want to add semantic signals without throwing lexical search away
FAQ
No. The whole stack runs on CPU. all-MiniLM-L6-v2 and the cross-encoder are small, fast models designed for CPU inference.
Qdrant runs locally in Docker, stays free, and supports dense plus sparse in one collection. You learn the primitives instead of renting them.
The architecture is. The sparse embedder is a simplified BM25-style hasher for teaching. For production, swap in SPLADE or a corpus-tuned BM25. Everything else maps directly to production systems.
Pricing
One subscription unlocks every paid course and workshop replay. Pick yearly or monthly.
Unlock with Pro
You save 47% with regional pricing
Billed annually. Cancel anytime.
Still deciding? Ask Param a question
Hybrid document search with Qdrant and Sentence Transformers
From $16/mo with Pro