Qdrant

🔍 𝐏𝐨𝐰𝐞𝐫𝐢𝐧𝐠 𝐋𝐚𝐫𝐠𝐞-𝐒𝐜𝐚𝐥𝐞 𝐍𝐋𝐏 𝐏𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬 𝐰𝐢𝐭𝐡 𝐐𝐝𝐫𝐚𝐧𝐭 𝐕𝐞𝐜𝐭𝐨𝐫 𝐒𝐞𝐚𝐫𝐜𝐡 In a recent production talk, Justin Miller from Zefr shared how their team processes millions of social media posts across TikTok, YouTube, and Instagram using an embedding-first architecture - with Qdrant as the vector search backbone. How Qdrant fits into the system: - Store and index embeddings generated from large-scale NLP pipelines - Support high-throughput vector ingestion from GPU-accelerated workloads - Enable fast similarity search across constantly changing, multi-platform content - Integrate cleanly with data stacks like Snowflake and GCS - Handle shard lifecycle and continuous updates without slowing retrieval The result is a production-grade system where vector search isn’t an afterthought - it’s a core primitive for understanding, organizing, and retrieving content at scale. A strong example of how Qdrant supports real-world, high-volume AI pipelines in production. 📺 Watch the full talk: https://lnkd.in/ghfWbGNc #Qdrant #VectorSearch #Embeddings #NLP #ProductionAI #SemanticSearch

🏭 𝐄𝐦𝐛𝐞𝐝𝐝𝐢𝐧𝐠-𝐁𝐚𝐬𝐞𝐝 𝐒𝐮𝐫𝐟𝐚𝐜𝐞 𝐃𝐞𝐟𝐞𝐜𝐭 𝐃𝐞𝐭𝐞𝐜𝐭𝐢𝐨𝐧 𝐰𝐢𝐭𝐡 𝐐𝐝𝐫𝐚𝐧𝐭 Modern factories generate thousands of defect images, but traditional rule-based or classification systems often miss context. This approach shows how embedding-based similarity search with Qdrant can unlock deeper insights from visual defect data. This is an amazing article by Abhinaya Pinreddy 👏 Key highlights: - Convert surface defect images into vector embeddings instead of relying only on labels - Use Qdrant vector search to instantly find similar historical defects - Retrieve contextual metadata like defect type, severity, and past resolution steps - Enable engineers to learn from previous cases, not start from scratch every time - Move from pure detection to experience-driven quality control A strong example of how vector search extend computer vision systems beyond classification - toward smarter, more explainable industrial AI. More details here: https://lnkd.in/gRAaPJZr #VectorSearch #Embeddings #ComputerVision #ManufacturingAI #IndustrialAI #Qdrant

Welcome to the Qdrant Distinguished Ambassador Program, Tarun R Jain! 🎉 We’re excited to onboard Tarun as a Distinguished Ambassador at Qdrant Tarun has been an incredible force in the community - consistently going above and beyond through: - 🌍 Global tech talks - 🐍 PyCons and developer conferences - 🎙️ Webinars and knowledge-sharing sessions - 💻 Meaningful open-source contributions From educating developers to strengthening the ecosystem, his impact has been truly outstanding. We’re grateful for everything he’s done and thrilled to continue this journey together. Welcome aboard, Tarun - well deserved! 👏

A very comprehensive guide, pushed away from the spotlight by the winter holidays:) Covers: + Dense vector search (how else?) + Keywords-based search + Obviously, BM25; + Sparse neural retrieval (SPLADE, miniCOIL, etc.). It's when you search based on keywords, but the retriever accounts for their meaning in the context of the query/documents. + Full-text indexing & filtering, which doesn’t break vector search (cause filterable HNSW/ACORN) + How to combine approaches 🌲 https://lnkd.in/dsWUqFwp

Is your Lucene-based vector search hitting a wall? If you are running vector search on Lucene (Elastic/OpenSearch) and struggling with latency or indexing costs, you aren't alone. We’ve seen teams make the switch and see immediate, massive gains. Just look at Sprinklr. After moving to Qdrant, they achieved: 90% faster write times, 80% faster latency, 2.5x higher RPS...all compared to their previous Lucene-based stack. !!We are currently offering a free technical POC comparison for teams looking to benchmark their current setup against Qdrant!! 📩 DM me if you want to apply. Let’s see if we can replicate these numbers for you. more info: https://lnkd.in/dbkaqyNA #VectorDatabase #Qdrant #Lucene #Elasticsearch #AI #MachineLearning #TechInfrastructure

I hope you still like my advanced design skills. Yes, we are hiring again! The Qdrant team opened several positions, and we will add more soon.  If you’re excited about vector search, AI infrastructure, and building developer-first products, this might be the perfect time to join us. 😎 We hire: - Developer Relations Engineers (EU/US) - Benchmark Engineer (Worldvide) - Solution Architects (EU/US/India) - Support Engineers (EU/US/India) - Account Executives (EU/US) - Cloud Engineers (EMEA) - DevOps/SRE (EMEA) - And more here ⤵️ QDRANT.JOIN.COM If you are interested, and you should be. 😉 Please do not write me directly! 𝐈'𝐦 𝐧𝐨𝐭 𝐭𝐡𝐞 𝐡𝐢𝐫𝐢𝐧𝐠 𝐦𝐚𝐧𝐚𝐠𝐞𝐫. Please apply via qdrant.join.com Also, for recruiters. You are welcome, but... we use HireBuffer as a Proxy service to gather all candidates through a single channel. Signup here for free: hirebuffer.com?ref=qdrant and share your candidates with us. #hiring #recruiting #jobs #careers

𝐁𝐞𝐭𝐭𝐞𝐫 𝐑𝐀𝐆 𝐒𝐭𝐚𝐫𝐭𝐬 𝐰𝐢𝐭𝐡 𝐁𝐞𝐭𝐭𝐞𝐫 𝐄𝐱𝐩𝐞𝐫𝐢𝐦𝐞𝐧𝐭𝐬 😉 RAG performance often comes down to one question: which chunking strategy actually works best? The Tigris Data RAG Lab shows how reproducible experimentation can make that decision data-driven - by running parallel chunking variants as Tigris bucket forks, and pairing each dataset with its own Qdrant vector index for clean, apples-to-apples evaluation. With fast similarity search and scalable indexing, Qdrant helps teams compare retrieval quality objectively and iterate on RAG systems with confidence - not guesswork. A solid example of how thoughtful data workflows + vector search lead to better AI outcomes. More details here: https://lnkd.in/gSdzUb4F #RAG #VectorSearch #LLMOps #AIInfrastructure #Qdrant #RetrievalAugmentedGeneration

https://lnkd.in/gi8-veCc Here’s my talk on generating embeddings at scale. Huge thanks to Ray Summit for the opportunity to present on the topic. Description: At Ray Summit 2025, Justin Miller from ZEFR shares how his team built a production-grade, multi-platform NLP pipeline using Ray and GPU acceleration to process millions of social media posts across TikTok, YouTube, and Instagram. He begins by describing the challenges of handling massive, fast-changing content streams across multiple platforms—each with unique data formats, ingestion patterns, and quality constraints. To meet these demands, ZEFR engineered a robust distributed pipeline that uses Ray to orchestrate scalable embedding generation, GPU-heavy processing, and high-throughput vector search ingestion. Justin walks through the architecture step-by-step: Snowflake → Ray ingestion: Retrieve rows for each platform with consistent batch scheduling Cleaning, chunking, and preprocessing: Normalize and prepare multimodal content at scale Distributed embedding generation: Use Ray Actors to shard GPU inference tasks across the cluster High-throughput writes: Send results to Google Cloud Storage (GCS), Qdrant for vector search, and back to Snowflake for analytics and pipeline tracking Shard lifecycle management: Delete stale shards, manage multi-platform ingestion, and maintain healthy storage footprints He also shares practical, real-world guidance for operating Ray in production—covering deployment patterns, debugging tips, failure recovery, throughput tuning, and cost management. Whether you’re processing large multi-source datasets, running GPU-heavy inference pipelines, or building modern vector-search–backed systems, this talk provides both code-level insights and actionable advice for running Ray at scale. #ray #anyscale #embeddings

Flipkart’s Trust & Safety team has made significant advancements by replacing slow batch-based similarity checks, which took up to 9 hours, with a real-time vector search pipeline utilizing Qdrant. This transformation involved indexing high-dimensional embeddings and executing online similarity searches, which reduced detection latency to less than 1 minute. This shift enables proactive fraud prevention rather than relying on post-hoc cleanup. This case exemplifies how vector search solutions are evolving from "AI experiments" to essential infrastructure for real-time systems. Check it out: https://lnkd.in/dsi38HKn Thanks to Sourabh Sarkar for contributing to this content piece!