Live coding a quick & easy semantic search with Qdrant from scratch
summary
Live codes a working semantic search built in Rust end to end: an rust-bert embedding pipeline using the miniLM L12 v2 model (384‑dim, cosine similarity) via the tch/libtorch crate, inserts vectors and JSON payloads into Qdrant over gRPC, and performs searches that return semantically related docs (example finds “quantization” docs from partial query). Implementation chunks upserts (batches of ~100), uses a JsonL input file, relies on HNSW indexing and supports product/quantization, and it works live.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 2:15Semantic search explained
- 7:04Qdrant client setup
- 14:28Reading data & embeddings
- 25:09Inserting data with batching
- 28:26Implementing search
- 30:20Q&A on Qdrant
- 37:10Performance & scalability
- 40:23Distance metrics & advanced search
- 43:12Use cases & limitations
- 45:25Closing remarks
related talks
49mHybrid search based on OpenSearch and Qdrant
32mBudget Friendly Semantic Search With Neural Sparse Search
30mSemantic Search and Neural Information Retrieval
55mDive into the Future: .NET and AI!
31mVerifAI: Biomedical QA with Verifiable Answers
26mAdvancing Multi-Modal Search Capabilities in Search Pipeline
32mRetrieval Augmentation and Semantic Search at Scale
33mA Practical Approach To Semantic Search