Qdrant

miniCOIL: on the Road to Usable Sparse Neural Retrieval - Qdrant Introducing miniCOIL, a lightweight sparse neural retriever capable of generalization.

This is really neat. I really like the way that Qdrant focuses on simplicity and performance while still trying to combine it with the insights from heavier machine learning innovations for meaning extraction and semantic search.

An open source engineer from @llamaindex.bsky.social shares trial-and-error-learned lessons.

For anyone implementing textual RAG, check the blog by @cle-does-things.bsky.social!
✅ useful tips, from chunking to evals-related;
✅ projects applying each tip in the wild.

👉 qdrant.tech/blog/hitchhi...

Qdrant + DataTalks.Club: Free 10-Week Course on LLM Applications - Qdrant Gain hands-on experience with RAG, vector search, evaluation, monitoring, and more.

Free 10-week course! Learn RAG, vector search, hybrid search, evaluation, and more. Hands-on lessons from Qdrant’s experts.

Learn more 👉 qdrant.tech/blog/datatal...

@mrscoopers.bsky.social and Kacper Łukawski will be there June 16-17 in Berlin, so if you're into vector search and want to chat over coffee, come find us! Plus Jenny's giving a talk about miniCOIL - our new sparse retrieval model that's pretty cool 🚀

Coffee + vector databases = perfect combo ☕

☕ Guess what? We're sponsoring the coffee breaks at @berlinbuzzwords.de 2025!

⏰ 𝐎𝐧𝐞 𝐰𝐞𝐞𝐤 𝐭𝐢𝐥𝐥 𝐁𝐞𝐫𝐥𝐢𝐧 𝐁𝐮𝐳𝐳𝐰𝐨𝐫𝐝𝐬
I'll really try to make my talk (on the 17th at 11:10 am in Kesselhaus) fit this "𝘏𝘮𝘮𝘮𝘮, 𝘐 𝘴𝘩𝘰𝘶𝘭𝘥 𝘵𝘳𝘺 𝘵𝘩𝘪𝘴 𝘳𝘯" vibe of @berlinbuzzwords.de!
To TLDR, it's about making 𝐬𝐩𝐚𝐫𝐬𝐞 𝐩𝐚𝐫𝐭 𝐢𝐧 𝐡𝐲𝐛𝐫𝐢𝐝 𝐬𝐞𝐚𝐫𝐜𝐡𝐞𝐬 as 𝐫𝐞𝐥𝐢𝐚𝐛𝐥𝐞 as BM25, but 𝐬𝐞𝐦𝐚𝐧𝐭𝐢𝐜𝐚𝐥𝐥𝐲 𝐚𝐰𝐚𝐫𝐞 (so 𝐛𝐞𝐭𝐭𝐞𝐫 than BM25)

Bavaria, Advancements in SEarch Development (BASED) Meetup · Luma Meetup Description: Search is evolving fast. From new algorithms and tools to AI-driven solutions, the field is constantly shifting. BASED Meetup is where…

Join for:

✅ practical advice on vector search in daily #AI developer’s life from our & @llamaindex.bsky.social beloved star Clelia Astra Bertelli
✅ research insights on the role of multimodality in industrial #RAG from Monica Riedler, amazing PhD at @tum.de‬

Register here
👉 lu.ma/cyelt1n6

🥨 Come to the Bavarian Search Meetup!

The 2nd meeting of 𝐁avaria, 𝐀dvancements in 𝐒𝐞arch 𝐃evelopment meetup, co-organized by our @mrscoopers.bsky.social, is happening in #Munich on the 12th of June!

👇

spot art. text reads: Announcing .NET AI Chat Web App Template

Create smart chat apps with custom data and enjoy easy setup in Visual Studio, VS Code, or CLI.

Preview 2 of the .NET AI Template is here, and now it's easier than ever to build cloud-native #AI apps with #dotNETAspire and #Qdrant vector DB integration. Learn. https://msft.it/63326S1eg4

Yeah! @qdrant.bsky.social 😍 .NET AI :)

🚀 Building a RAG API with .NET, Semantic Kernel, Phi-3, and Qdrant: Enrich Your E-commerce Experience Learn how to build a powerful RAG (Retrieval-Augmented Generation) API using .NET, Microsoft Semantic Kernel, Phi-3, and Qdrant. Combine your private e-commerce data with LLMs to create smarter, gr…

🚀 Build a real RAG API with .NET 8, Semantic Kernel, Phi-3, and Qdrant!

🎯 Enrich LLMs with real product data
⚙️ Local ONNX inference with Phi-3
🔥 Full API endpoints ready!

Dive into the architecture 👉 wp.me/p29SK-Yu

#DotNet #SemanticKernel #RAG #Qdrant #Phi3 #AI #MachineLearning #ONNX

Want to see MCP in action? Learn how to orchestrate OpenAI agents using MCP, AugmentCode, and Qdrant.

📅 April 29 @ 11 am ET

🔗 Save your spot: try.qdrant.tech/mcp-agent-in...

Relevance Feedback in Informational Retrieval - Qdrant Relerance feedback: from ancient history to LLMs. Why relevance feedback techniques are good on paper but not popular in neural search, and what we can do about it.

The article explores the gap between theory and practice — and why relevance feedback hasn't made it yet into neural search at scale.

If you're building or working with retrieval systems, it’s worth a read.

👉 qdrant.tech/articles/sea...

Relevance feedback helps search systems iteratively improve results toward relevance. It’s been studied for 60+ years — yet remains rare in modern neural search.

We explored the field to understand why — and gathered this summary of methods proposed over the years.

⬇️

MCP week: Learn to use LlamaIndex to prep docs for Claude with a pre-built MCP server for @qdrant.bsky.social, using @angular.dev docs. Set up vector DB, process AI-friendly docs, configure MCP server, implement RAG pipeline:
https://www.aiboosted.dev/p/building-your-own-rag-system-typescript

Large Scale Search - Qdrant Qdrant is an Open-Source Vector Database and Vector Search Engine written in Rust. It provides fast and scalable vector similarity search service with convenient API.

🧵 Tutorial: qdrant.tech/documentatio...
📁 Code: github.com/qdrant/laion...

Large Scale Search - Qdrant Qdrant is an Open-Source Vector Database and Vector Search Engine written in Rust. It provides fast and scalable vector similarity search service with convenient API.

We documented everything: memory estimates, observed metrics, config breakdowns, and open-sourced all the scripts.

🧵 Tutorial: qdrant.tech/documentatio...
📁 Code: github.com/qdrant/laion...

Here's what we ran:
✅ Stored vectors as FLOAT16 instead of FLOAT32
✅ Used binary quantization to keep compressed vectors in RAM
✅ Kept full-precision vectors on disk for query-time rescoring
✅ Tuned HNSW for lower memory
✅ Enabled async disk I/O with io_uring to parallelize reads

🚧 We ran vector search over 400M CLIP embeddings from LAION-400M — using Qdrant and just 64GB RAM.

We wanted to see how far we could push Qdrant with minimal hardware and how much we could squeeze out of quantization, indexing, and async I/O.

Built for Vector Search - Qdrant Why add-on vector search looks good — until you actually use it.

Our opinion on this qdrant.tech/articles/ded...

Let us know how it goes. 🍿

Built for Vector Search - Qdrant Why add-on vector search looks good — until you actually use it.

qdrant.tech/articles/ded...

A lot of people ask us: “𝗪𝗵𝘆 𝘀𝗵𝗼𝘂𝗹𝗱 𝗜 𝘂𝘀𝗲 𝗮 𝗱𝗲𝗱𝗶𝗰𝗮𝘁𝗲𝗱 𝘃𝗲𝗰𝘁𝗼𝗿 𝗱𝗮𝘁𝗮𝗯𝗮𝘀𝗲 when I can simply add a vector plugin or extension to my existing database?”

We wrote an article addressing this very common question.

Article: qdrant.tech/articles/ded...

Optimizing Memory for Bulk Uploads - Qdrant Efficient memory management is key when handling large-scale vector data. Learn how to optimize memory consumption during bulk uploads in Qdrant and keep your deployments performant under heavy load.

If you're working with large-scale vector ingestion, you should know how to manage memory efficiently.

👉 Learn the best practices for optimizing memory in Qdrant: qdrant.tech/articles/ind...

Talks include:

✅ Scaling AI agents with vector search and RAG
✅ AI agents for workflow automation
✅Strategic value of AI agents in the enterprise
✅ AI performance monitoring and impact measurement
✅ Orchestrating AI agents with low-code automation

Qdrant AI Builders: Agents in the Enterprise @ AWS Loft with Arize, CrewAI, and n8n · Luma Join Qdrant, Arize, CrewAI, n8n, and AWS for an evening event tailored for executives and business leaders to explore the strategic value, possibilities, and…

Qdrant AI Builders: Agents in the Enterprise @ AWS Loft

With @arize.bsky.social, CrewAI, and @n8n.io 🚀

An evening for executives and technical leaders to explore AI agents and RAG in the enterprise.

🔗RSVP: lu.ma/etufeex8?tk=...

📅 Monday, February 10

Regions are just logical boundaries within the bitmask, the value can span more than one region by design. That’s why the gaps include leading and trailing, so that we can combine them when looking for a large gap.

We use u32 as internal ID, whose sequence is unique to each segment.

Bustle benchmark was run single-threaded, because we use our segments single-threaded.

Code: github.com/qdrant/qdran...

(not yet on master because of the rename, otherwise just look for blob_store )