
@zobo_07 great work!

High-performance Rust-based vector search engine. discord.com/invite/qdrant

@zobo_07 great work!


What if someone searches in Hindi but the answer exists in an English document? How do you bridge that gap? Akshat explored this using 2,066 Indian government schemes, comparing BM25, SPLADE, and BGE-M3 with Qdrant Hybrid Search. The experiment shows how lexical, sparse, and multilingual dense retrieval can complement each other to tackle vocabulary and language mismatches. A great practical example of hybrid search beyond traditional keyword matching.

Read the full breakdown by @itsclelia here: qdrant.tech/articles/embeddi…

Have you ever generated 32.8 million embeddings in 35 minutes? We did, using the potion static embedding models from @minishlab, that we recently integrated in Fastembed (our framework for local embeddings inference). In our latest article in collaboration with Minish, we dive into what static embeddings are, what optimizations we introduced in Fastembed to make them go brr, and what tradeoffs they carry in terms of retrieval quality. Link below👇


Qdrant 1.20 has brought plenty of updates! Join us for Office Hours on Thursday, November 19, for an open discussion with the Qdrant team. Bring your questions, share your experiences, and let's talk about what's new and how it can help your workloads.

You know, the fastest graph indexes take the longest to build. HNSW and Vamana, for example, spend most of their build time searching to find neighbors. This leads to random memory access, cache misses, and idle CPU cores waiting on DRAM. With a billion vectors, building can take hours or even days. Which is annoying, right? PiPNN is a new ultra-scalable, graph-based nearest-neighbor indexing approach that entirely removes the search step from graph construction.

How small can we go with the hardware we run our embedding models on? Turns out, pretty small: in my latest blog post at @qdrant_engine, I took potion-base-2M by @minishlab and, with some no_std Rust 🦀, I made it run on a ESP32 microcontroller with 16MB of memory. And it ran pretty fast as well! It embedded a dozen tokens in approximately 1.2ms, meaning a throughput of 830 embeddings/s (assuming very short texts). Of course, this is far from the 11-15k embeddings/s that I saw static models produce on my Mac with 48GB of unified memory (more on that very soon 👀) , but it's pretty impressive considering it uses 3000x less memory. Also, it's proof that we can start thinking about running search 𝘵𝘳𝘶𝘭𝘺 on the edge. Link to the blog below👇

(3/3) And with Qdrant’s MCP integrations, AI agents can: → Manage and monitor Qdrant Cloud clusters. → Inspect cluster health and performance. → Search official documentation and retrieve code examples. → Interact with Qdrant using natural language. MCP provides access to tools, while Agent Skills provide the technical knowledge to use them effectively. Release: github.com/qdrant/qdrant/rel… AI agents: qdrant.tech/llms.txt MCP: qdrant.tech/documentation/qd…

(2/3) Working with Qdrant through AI agents is getting easier. You can now give your coding agent a single instruction: Set up Qdrant with qdrant.tech/llms.txt Your agent gets a starting point to discover Qdrant documentation, install Agent Skills, and use the Cloud CLI. No more hunting through multiple resources to get started.

(1/3) Qdrant 1.20.1 is here, bringing some serious performance improvements. A few highlights: → 50% fewer reads during HNSW graph search on cold storage. → ~2× faster Turbo8 and Turbo16 quantization. → Faster HNSW graph construction with PiPNN. → Faster filtering, batch searches, shard transfers, and record streaming. But there’s another update worth talking about. 🧵

Search Mythbusters. Guess the Query. BM25 or Vectors? 11 days until Qdrant and @ClickHouseDB's London Search Week Pub Quiz! Nov 24, 6:30 PM GMT at Core Bar. Bring a team or come solo. Food, first drink, and prizes included. RSVP: luma.com/londonsearchweek_20…

Read the paper we presented here: dl.acm.org/doi/epdf/10.1145/…

Qdrant is at CIKM Rome: @krotenWanderung talked about vector index-native relevance feedback in production systems at Industry Track and, if you're curious about that or want to know how Qdrant can help your vector search in production, you can find her and get answers to any of your questions!

BM25 or Vectors? Time to assemble your team. 3 weeks until Qdrant and @ClickHouseDB's London Search Week Pub Quiz! Nov 24, 6:30 PM GMT at Core Bar. Food, first drink, and prizes on us. Come solo or bring up to 6. Limited spots. RSVP: luma.com/londonsearchweek_20…