Software Engineering Radio - the podcast for professional software developers SE Radio 691: Kacper Łukawski on Qdrant Vector Database
Oct 22, 2025
Kacper Łukawski, Senior Developer Advocate at Qdrant who focuses on vector databases and similarity search. He explains vector databases, embeddings, and when to use semantic search. They cover Qdrant’s Rust-based design, performance benchmarking and precision trade-offs, deployment options and client libraries, and how vector search supports retrieval-augmented generation and real-world applications.
AI Snips
Chapters
Transcript
Episode notes
Benchmark On Affordable Reproducible Hardware
- Run benchmarks on affordable, reproducible hardware so others can validate results.
- Avoid huge cloud instances because typical users run vector search on modest VPS instances.
Benchmark All Relevant Metrics
- Benchmarks must report throughput, latency (P95/P99), CPU, memory, and indexing time to reflect operational trade-offs.
- Indexing time matters when data changes frequently because it builds helper data structures for fast search.
Qdrant Benchmark Headlines
- In Qdrant benchmarks, indexing 1M OpenAI embeddings took ~24–25 minutes.
- That setup yielded ~3–4 ms average search latency, ~1200 QPS, and ~0.99 precision.