TechPrufer
Get started
Qdrant
Qdrant logoQdrant · Feature trail

Agentic Retrieval Systems

17 posts · Jan 2024 – Oct 2026

What this trail covers

Qdrant continues to be a core component for agentic retrieval systems and RAG evaluation, enhancing its capabilities for advanced retrieval, performance optimization, enterprise scaling, and deployment flexibility. This post highlights the use of Jev, a decision model, for various classification tasks within retrieval systems, including reranking, reducing repetitive answers via Maximal Marginal Relevance (MMR) guided by Jev scores, and query understanding through taxonomy routing. Experiments show Jev's effectiveness in improving relevance and diversity, often outperforming traditional methods and cross-encoders, while offering a more constrained and cost-effective alternative to generative LLMs for classification tasks.

All Qdrant trails
  1. Jevjitsu: Or, How We Tried Generalized Classifiers on Everything - QdrantThis post introduces experiments with Jev, a decision model, for several classification tasks within retrieval systems. It details Jev's application in reranking search results, demonstrating that Jev Score and Jev Choice (1 request per query) offer significant nDCG@10 lift over baseline embedding models and comparable or better results than local cross-encoders. It also explores using Jev to guide Maximal Marginal Relevance (MMR) for reducing repetitive answers and increasing product variety, showing that combining Jev with MMR can improve relevance while maintaining diversity. Furthermore, the post describes using Jev for query understanding and taxonomy routing, where Jev categorizes products and scores them against queries to filter or boost results, demonstrating its utility in improving search relevance on datasets like Amazon-C4.
  2. SHIFTing Languages in Multilingual RAG - QdrantThis post introduces and evaluates the SHIFT method for mitigating language bias in multilingual RAG systems. It demonstrates that the SHIFT method, by treating languages as offsets, can significantly improve cross-language retrieval recall without requiring additional training or translation pipelines. The post provides empirical evidence using the `multilingual-e5-small` model and the XRAG dataset, showing improvements in recall@10 for both same-language and cross-language scenarios. It also details findings on offset convergence, dataset dependency, and the impact of SHIFTing vectors on Qdrant's HNSW index.
  3. Qdrant and Minima Deliver 2.92x More Agentic RAG Tasks per GPU-Hour - QdrantThis post details a benchmark demonstrating a 2.92x increase in successful agentic RAG tasks per GPU-hour by combining Qdrant's hybrid search, payload filtering, and reranking with Minima's LLM weight and KV cache compression. The joint solution reduced median task latency from 21.3s to 7.7s and cut GPU cost per 1,000 successful tasks by 65%. Key technical contributions include: Qdrant's hybrid search (dense + BM25 + RRF fusion) improving first-pass evidence sufficiency from 72% to 87% and reducing context tokens by 56%; Minima's NVFP4 W4A4 weight compression and FP8 KV cache reducing model size and increasing resident sessions by 8.7x; and the combined effect leading to 3,158 successful tasks/GPU-hour compared to 1,081 baseline.
  4. How GoPerfect Built an Agentic Recruiting Workforce with Qdrant Cloud - QdrantThis post details how GoPerfect built an agentic recruiting workforce by leveraging Qdrant's hybrid search and multi-vector capabilities. They moved from a single dense embedding per candidate to a structured set of vectors, enabling more granular retrieval. This, combined with LLM orchestration, achieved near-perfect match accuracy (99.993%) compared to the industry baseline of 30%. The predictable retrieval latency of Qdrant was crucial for the agent's ability to run multiple parallel sub-queries and reason over the results within an interactive budget, completing complex searches in under 10 seconds.
  5. How Data Graphs Built a True Hybrid Graph RAG Platform - QdrantThis post details how Data Graphs built a 'True Hybrid Graph RAG Platform' by integrating Qdrant with their proprietary graph database engine and an Agentic AI layer. Key technical contributions include: 1. Real-time embedding from the graph into Qdrant collections with rich metadata for filtering. 2. A schema-first retrieval approach where the AI agent uses the graph's schema to decide between graph queries (GQL/Cypher) and vector retrieval from Qdrant. 3. Blended retrieval and reasoning, executing graph and vector queries in parallel and citing all data points back to their source in the knowledge graph for verifiable provenance. The post highlights Qdrant's Hybrid Cloud deployment, payload filtering capabilities, and Terraform support as critical factors for their production environment. It emphasizes that combining vector search with graph retrieval, orchestrated by an agent, is crucial for building advanced AI applications beyond basic RAG.
  7. Qdrant Skills for AI Agents - QdrantThis post introduces 'Qdrant Skills,' a new open-source initiative to encode Qdrant's production knowledge into a format navigable by AI agents. It addresses the 'agent gap' where agents can call APIs but lack the diagnostic and decision-making expertise for complex system tuning. Skills are organized by problem (e.g., 'What to do if search quality is bad?') rather than by feature, providing imperative guidance, 'use when' triggers, and crucial 'what NOT to do' advice, linking directly to documentation. The post demonstrates significant improvements in agent performance (96% assertion pass rate vs. 65%) when using skills, highlighting their value in navigating Qdrant's engineering tradeoffs like memory vs. latency and recall vs. throughput. It also mentions the complementary qcloud-cli for agentic infrastructure management.
  8. How My AskAI Built Self-Improving Support Agents - QdrantThis post details how My AskAI built self-improving AI customer support agents using Qdrant. It highlights the transition from a general 'chat with your data' product to a specialized customer support solution, emphasizing the unique retrieval requirements of messy, unstructured, and constantly changing support data. The adoption of embeddings and RAG, followed by experimentation with hybrid search, is discussed. My AskAI chose Qdrant Cloud for its scalability, predictable pricing, integration ecosystem, and developer experience. The post details the benefits realized after migrating to Qdrant Cloud, including simplified scaling and management, and the ability to easily enable hybrid search per collection. A key future development is the self-learning mechanism where human agent responses are indexed as 'self-learning articles' in Qdrant, allowing agents to answer questions correctly the next day.
  9. Two Approaches to Helping AI Agents Use Your API (And Why You Need Both) - QdrantThis post introduces two complementary approaches for enabling AI agents to effectively use APIs: SKILL.md for providing static, pre-computed knowledge about API usage (known unknowns) and a REPL-first approach for dynamic, interactive discovery of the API's current state (unknown unknowns). It details how these methods address distinct failure modes in agentic API interaction and proposes a combined workflow where agents first consult SKILL.md for guidance on correct API usage (e.g., deprecated methods, anti-patterns) and then use a REPL to explore the live environment (e.g., list collections, inspect schemas) before generating code. The post provides implementation details for a Qdrant REPL tool and discusses the broader applicability of this pattern to any API with deprecated methods or user-specific state.
  10. How Anima Health scaled clinical document intelligence with Qdrant - QdrantThis post details how Anima Health uses Qdrant as a core infrastructure component for their clinical document intelligence platform. Specifically, Qdrant serves as a retrieval layer within an agentic pipeline to narrow the search space for SNOMED codes, enabling high-confidence recommendations for clinicians. The platform leverages Qdrant's payload-based filtering to scope searches across different EHR systems and preserve patient privacy by working with embedded representations of documents. Deployment control (self-hosted, UK residency) and cost predictability were key factors in choosing Qdrant. The article highlights Qdrant's role in scaling the system and enabling future agentic workflows, including modeling patient state and temporal relevance.
  11. How Kakao Built an AI-Powered Internal Service Desk with Qdrant - QdrantThis post details how Kakao built an AI-powered internal service desk using Qdrant as a core component of their RAG system. It highlights the use of Qdrant's hybrid search (dense and sparse vectors with RRF) and named vectors to address the challenge of searching complex internal knowledge bases that require both semantic understanding and exact term matching. The implementation involved an automated indexing pipeline, metadata filtering, and a semantic cache. Qdrant was deployed on Kubernetes for high availability, demonstrating its operational fit for enterprise internal services. The post emphasizes the benefits of improved search quality, faster response times, and reduced support load.
  12. How Xaver scaled personalized financial advice with Qdrant - QdrantThis post details how Xaver, a financial advisory platform, scaled its personalized financial advice service by implementing a two-tier knowledge engine using Qdrant. The first tier uses a condensed knowledge base (CKB) for common queries, enabling near-instant recall. The second tier uses a full knowledge base for more complex queries. Qdrant was chosen for its Rust-based core, low latency, developer simplicity, and flexible deployment, allowing Xaver to focus on its application layer rather than infrastructure.
  13. How TrustGraph built enterprise-grade agentic AI with Qdrant - QdrantThis post details how TrustGraph built an enterprise-grade agentic AI platform by integrating Qdrant's vector search with Apache Pulsar for streaming and graph-native semantics (RDF/SPARQL). It highlights a retrieval process that goes beyond traditional RAG by expanding Qdrant-retrieved entities into a subgraph for LLM context, ensuring determinism and auditability. The architecture is designed for production, focusing on resilience, scalability, and data sovereignty, contrasting with typical demo-based approaches.
  14. How Fieldy AI Achieved Reliable AI Memory with Qdrant - QdrantFieldy AI migrated from Weaviate to Qdrant for their AI memory platform due to reliability issues with Weaviate (5xx errors). The migration involved reusing the existing vector schema, batching embeddings locally before ingestion, disabling indexing during bulk import, and updating API calls to use Qdrant's gRPC interface. They implemented hybrid search with BM25 and dense vector retrieval via Reciprocal Rank Fusion (RRF). The migration resulted in 100% reliable recall, reduced latency, virtually eliminated network errors, and a two-thirds reduction in infrastructure costs while scaling to tens of millions of embeddings.
  15. How OpenTable Reinvented Restaurant Discovery with Qdrant - QdrantThis post details how OpenTable built its Concierge AI Dining Assistant using Qdrant. Key technical contributions include the use of Qdrant for handling sparse embeddings and achieving high-precision filtering for restaurant discovery, enabling queries to narrow down results to a single restaurant out of over 60,000. The choice of Qdrant was driven by its ability to manage sparse embeddings effectively without significant HNSW graph quality degradation and its predictable performance for complex filtering requirements involving reviews, metadata, and structured data. The post highlights the ease of deployment with Qdrant Cloud and the operational stability achieved in production, which was crucial for the success of the Concierge rollout.
  16. How PortfolioMind Delivered Real-Time Crypto Intelligence with Qdrant - QdrantPortfolioMind implemented a system using Qdrant to model dynamic user curiosity in the crypto domain. User interactions (searches, pins, reads) are converted into semantic traces and then into multivector user-intent models. Diverse data (news, tokenomics, whale behaviors) is ingested with rich metadata. HDBSCAN clustering identifies user-specific micro-interests, creating a dynamic, multivector representation. Qdrant's features like fast, filterable searches, native multivector support, low-latency retrieval, and managed services were crucial. This resulted in a 70% latency drop, 58% interaction relevance increase, and 22% user retention rise.
  17. Qdrant for Research: The Story Behind ETH & Stanford’s MIRIAD Dataset - QdrantThis post details the creation and use of the MIRIAD dataset, a large-scale collection of medical Q&A pairs grounded in peer-reviewed literature. Qdrant was used for storage and RAG experiments, demonstrating its speed and scalability for handling millions of data points. The dataset aims to reduce hallucinations in medical AI applications and serves as a knowledge base for RAG and embedding model downstreaming. The MIRIAD Atlas, an interactive visualization tool, is also being migrated to Qdrant.
  18. How Pento modeled aesthetic taste with Qdrant - QdrantThis post details how Pento modeled aesthetic taste using Qdrant. It describes a system that converts user-art interactions into a semantic vector space. Key technical contributions include: 1. Interaction acquisition and conversion into signed weights (positive/negative affinity). 2. Artwork embedding using image encoders. 3. Interaction clustering with HDBSCAN to identify aesthetic themes, using medoids for cluster representation. 4. Scoring taste clusters by recency using an exponential decay function and keeping the top 50% of clusters. 5. User representation as multivectors (positive and negative). 6. Retrieval via Qdrant's Recommendation API, leveraging its scoring strategy for multivectors to find artists that overlap with positive preferences while avoiding negative ones. This demonstrates a novel application of vector databases for personalized, nuanced recommendation systems beyond traditional collaborative filtering.
  19. How Alhena AI unified its AI stack and improved ecommerce conversions with Qdrant - QdrantThis post details how Alhena AI unified its fragmented vector search infrastructure (previously using FAISS and Pinecone) onto Qdrant Cloud. Key technical contributions include: the successful migration of all traffic to a single backend, enabling hybrid search (dense and sparse embeddings) with metadata filtering and boosting for improved e-commerce conversions, achieving up to 90% reduction in retrieval latency, implementing multitenancy within shared collections for efficient scaling, and leveraging Qdrant Cloud for simplified global deployment and faster feature adoption. The architecture integrates Qdrant into a multi-stage agent pipeline involving query rewriting, safety evaluation, and hallucination detection.
  20. GraphRAG: How Lettria Unlocked 20% Accuracy Gains with Qdrant and Neo4j - QdrantThis post details Lettria's implementation of a GraphRAG system using Qdrant and Neo4j, achieving a 20-25% accuracy improvement over traditional RAG. Key technical contributions include: a custom ingest transaction mechanism to ensure atomicity between Neo4j (transactional) and Qdrant (non-transactional) by using snapshots and conflict resolution for rollback; payload flattening to convert nested JSON to flat key-value pairs for consistent indexing and filtering across both databases; and scaling to over 100 million vectors with sub-200ms P95 retrieval latency by implementing payload indexes on frequently filtered fields and using disk-based collections for cold data to manage memory pressure.
  21. ​​Introducing the Official Qdrant Node for n8n - QdrantIntroduces an official, team-supported n8n node for Qdrant, enabling direct integration of Qdrant's capabilities (hybrid search, reranking, filtering, batch operations) into n8n workflows without relying on HTTP nodes. Demonstrates a hybrid search example using Reciprocal Rank Fusion (RRF) within n8n.
  22. How Qovery Accelerated Developer Autonomy with Qdrant - QdrantThis post details Qovery's integration of Qdrant Cloud to power their AI-driven DevOps Copilot. Qovery selected Qdrant for its open-source credibility, performance, ease of use, and scalability, particularly for handling a large, continuously updated corpus of infrastructure data (approaching 500,000 vectors). The integration enabled Qovery to reduce reliance on specialized DevOps engineers by allowing developers to manage infrastructure tasks via natural language. Qdrant's real-time indexing and low-latency queries were crucial for delivering rapid and accurate responses, reducing infrastructure task execution time from hours/days to seconds. The Rust-based architecture of Qdrant provided the speed, accuracy, and low resource utilization required, with minimal operational overhead.
  23. How Tripadvisor Drives 2 to 3x More Revenue with Qdrant-Powered AI - QdrantThis post details how Tripadvisor is using Qdrant to power its AI Trip Planner, leading to a 2-3x revenue uplift from engaged users. It highlights the use of Qdrant to build a user graph for personalized travel experiences by representing user engagement with hotels, attractions, food, and travel styles. The article emphasizes Qdrant's suitability for real-time, unstructured data to power conversational AI, search augmentation, and recommendation engines, moving beyond traditional filters to conversational search.
  24. How Pariti Doubled Its Fill Rate with Qdrant - QdrantThis post details Pariti's successful implementation of Qdrant Cloud to significantly improve their talent matching process. Key technical contributions include: integration of Qdrant Cloud for zero-ops deployment and low latency; use of an intuitive Python SDK for embedding candidate data and role requirements; achieving millisecond query response times (22-40ms) with 0% downtime; reducing candidate vetting time by 70% (from 4 to 1 minute); and improving the success rate of 'database finds' from 6% to 24%. The post also mentions Pariti's exploration of Qdrant's GPU-assisted indexing and vector quantization features for future scaling.
  25. How Lyzr Supercharged AI Agent Performance with Qdrant - QdrantThis post details Lyzr's migration from Weaviate and Pinecone to Qdrant to address scaling limitations and performance bottlenecks in their AI agent platform. Key technical contributions include: quantitative improvements in query latency (>90% reduction), indexing speed (2x faster), and infrastructure cost reduction (30%). It highlights Qdrant's ability to sustain high throughput (>250 QPS) under heavy load (>100 concurrent agents, >1000 QPM) and its efficient handling of live updates without downtime. Case studies for NTT Data and NPD further illustrate Qdrant's impact on retrieval accuracy and low-latency performance for agent-driven applications.
  26. Vibe Coding RAG with our MCP server - QdrantThis post details the integration of Qdrant with the Model Context Protocol (MCP) to enable 'vibe coding' using AI assistants. It showcases how the `mcp-server-qdrant` implementation acts as a semantic memory layer, allowing AI agents to store and retrieve code snippets and documentation, perform semantic searches across codebases, and find relevant context for generating new code. The post provides specific examples of configuring Claude Code with the MCP server, including environment variables for tool descriptions and Qdrant connection details. It highlights the use of DaisyUI component documentation as a knowledge base for the AI to generate contextually relevant frontend code, demonstrating a practical application of RAG for AI-assisted development.
  27. How Deutsche Telekom Built a Multi-Agent Enterprise Platform Leveraging Qdrant - QdrantThis post details how Deutsche Telekom built and scaled their multi-agent enterprise platform (LMOS) using Qdrant. It highlights the challenges of scaling enterprise AI agents, including tenancy, horizontal scaling, and non-deterministic collaboration, framing them as distributed systems problems. The adoption of Qdrant is justified by its simplicity, Rust-based efficiency, memory management, and operational fit within their PaaS-first approach and multitenancy requirements, contrasting it with more complex alternatives. The platform now powers over 2 million conversations, reducing agent development time significantly.
  28. How to Build Intelligent Agentic RAG with CrewAI and Qdrant - QdrantThis post details the integration of Qdrant with CrewAI for building agentic RAG systems. It introduces a custom `QdrantStorage` class to enable CrewAI's entity and short-term memory to leverage Qdrant for storing and retrieving embeddings. The post also discusses strategies for loading knowledge into Qdrant, including semantic chunking and contextualization of Markdown notes using LLMs, and provides example YAML configurations for agent roles and tasks within CrewAI.
  29. Voiceflow & Qdrant: Powering No-Code AI Agent Creation with Scalable Vector Search - QdrantThis post details Voiceflow's integration of Qdrant for their no-code AI agent creation platform. It highlights their migration from Pinecone due to Qdrant's scaling capabilities, robust infrastructure support (including Private Link for SOC2 compliance), and responsive team. Voiceflow utilizes Qdrant's Product Quantization and chunking search for efficient data processing and retrieval. They also leverage Qdrant's isolated data management for multitenancy and anticipate future innovation in hybrid search.
  30. Building a Facial Recognition System with Qdrant - QdrantThis post details the architecture and implementation of a facial recognition system (Twin Celebrity app) using Qdrant. It covers the offline pipeline for generating and storing celebrity embeddings (using FaceNet and MTCNN) in Qdrant, and the online phase for querying user selfies against these embeddings. It also discusses deployment options and testing strategies, emphasizing Qdrant's role in enabling practical AI applications.
  31. Optimizing ColPali for Retrieval at Scale, 13x Faster Results - QdrantThis post details an optimization strategy for handling high-dimensional vector data generated by ColPali, specifically for visually rich PDFs. It introduces a two-stage retrieval process combining pooling (mean and max pooling on grid rows) to reduce the number of vectors per page from 1030 to 38, and a reranking stage using original high-resolution embeddings. This approach achieved a 13x speed improvement in retrieval time while maintaining high accuracy (NDCG@20 = 0.952, Recall@20 = 0.917 with mean pooling). The implementation involved creating a custom dataset and evaluating performance using NDCG@20 and Recall@20 metrics.
  32. Best Practices in RAG Evaluation: A Comprehensive Guide - QdrantThis post provides a comprehensive guide to evaluating RAG systems, detailing best practices for accuracy and quality. It covers common pitfalls in data ingestion, embedding, retrieval, and LLM generation, offering concrete solutions. Frameworks like Ragas, Quotient AI, and Arize Phoenix are introduced for systematic evaluation. Specific metrics like Precision@k, MRR, and DCG/NDCG are highlighted for retrieval assessment. The importance of chunking strategies and selecting appropriate embedding models (referencing MTEB) is emphasized.
  33. Empowering QA.tech’s Testing Agents with Real-Time Precision and Scale - QdrantThis post details QA.tech's successful implementation of Qdrant for their AI testing agents. It highlights Qdrant's ability to handle high-velocity embedding lookups generated by frequent user-like actions on web pages, overcoming scalability limitations previously encountered with pgvector. Key technical contributions include demonstrating how Qdrant's batch operations reduce network overhead, its efficient architecture optimizes CPU load for embedding processing, and its support for multiple embeddings per data point allows for diverse analytical needs (e.g., high-velocity action embeddings vs. deeper semantic search embeddings). The post emphasizes Qdrant's role in enabling real-time precision and scale for complex, multi-step AI agent workflows.
  34. Data-Driven RAG Evaluation: Testing Qdrant Apps with Relari AI - QdrantThis post introduces a data-driven RAG evaluation framework by integrating Qdrant with Relari AI. It details how to use Qdrant for storing and retrieving data for RAG applications and leverages Relari to evaluate performance through Top-K parameter optimization and auto-prompt optimization. The post provides practical steps and code examples for setting up the environment, defining a golden dataset, building a RAG app with Qdrant and LangChain, logging retrieval results, and submitting them to Relari for analysis.
  35. Kern AI & Qdrant: Precision AI Solutions for Finance and Insurance - QdrantThis post details Kern AI's use of Qdrant for building precision AI solutions in the finance and insurance sectors. It highlights Qdrant's multi-vector storage, easy setup, open-source nature, and rapid progress as key factors in Kern AI's choice. The integration enables Kern AI to achieve under 1% hallucination rates in RAG chatbots, significantly reducing customer service response times (e.g., Markel Insurance SE from 5 minutes to 30 seconds). Use cases include claims management and similarity search for incident handling.
  36. Developing Advanced RAG Systems with Qdrant Hybrid Cloud and LangChain - QdrantThis post details the integration of Qdrant Hybrid Cloud with LangChain to develop advanced RAG systems, specifically for an employee onboarding use case. It highlights the flexibility of Qdrant Hybrid Cloud's Kubernetes-native architecture for deployment in any environment, enabling users to leverage LangChain's framework for context-aware reasoning and semantic search. The integration aims to address LLMs' lack of user-specific knowledge by infusing them with relevant context from Qdrant's vector database, leading to improved retrieval accuracy and query speeds. The post also points to a tutorial for building a RAG system with Qdrant Hybrid Cloud, LangChain, and Cohere embeddings, and directs users to Qdrant's documentation for Hybrid Cloud deployment.
  37. Advancements and Challenges in RAG Systems - Syed Asad | Vector Space Talks - QdrantThis post, a transcript of a podcast episode, discusses the advancements and challenges in Retrieval Augmented Generation (RAG) systems, with a particular focus on multimodal RAG and the use of Qdrant. Key technical points include the challenges of accent variations and transcription accuracy in multimodal RAG, the cost implications of cloud deployment (specifically AWS), and the use of Qdrant for efficient vector search in a family tree app that utilizes video interactions. The speaker highlights Qdrant's commendability in providing excellent search results for accuracy-critical applications and mentions its ease of use compared to other vector databases. The discussion also touches upon the privacy paradox in AI and the preference for local deployment solutions like Ollama.
  38. Building Search/RAG for an OpenAPI spec - Nick Khami | Vector Space Talks - QdrantThis post details how to build search/RAG over an OpenAPI specification using Trieve and Qdrant. It highlights Qdrant's group-based system for organizing search results by API tags, enabling more relevant and structured retrieval compared to traditional single-route searches. The implementation involves loading the OpenAPI spec, creating groups for each tag, and inserting individual API paths as points within their respective groups. The use of Bun for scripting and Trieve's SDK is also demonstrated.
  39. Iveta Lohovska on Gen AI and Vector Search | Qdrant - QdrantThis post features an interview with Iveta Lohovska discussing the importance of trustworthiness and explainability in Generative AI, particularly in high-stakes use cases. It highlights the role of vector databases in providing control and explainability over data sources and foundational datasets. The discussion touches on implementing Gen AI in enterprise contexts like critical infrastructure, government, and intelligence agencies, emphasizing the need for accuracy and source credibility. A specific example of a climate agent is provided, stressing the importance of linking AI-generated answers to exact sources for traceability and ensuring scientific accuracy. The post also briefly mentions on-premises solutions for security and control, and the maturity of Gen AI models.
  40. How to meow on the long tail with Cheshire Cat AI? - Piero and Nicola | Vector Space Talks - QdrantThis post highlights the integration of Qdrant as the native database for the Cheshire Cat AI open-source framework. It details how Cheshire Cat utilizes Qdrant for its RAG system, including the use of scalar quantization to improve accuracy and performance while saving memory. The post also mentions the framework's support for various Qdrant deployment modes (file-based, container, cloud) and its plugin system, which allows for extending functionality. The "drunken cut effect" is described as an example of exploring advanced Qdrant features like aliases with multiple embedders.
  41. VirtualBrain: Best RAG to unleash the real power of AI - Guillaume Marquis | Vector Space Talks - QdrantThis post highlights VirtualBrain's use of Qdrant as a core component for their AI assistant for knowledge workers, specifically for drafting sales proposals using past reports. It emphasizes Qdrant's scalability, speed, and low latency as crucial for handling large datasets and parallel requests. The discussion touches on the importance of fresh data, techniques to boost scalability and speed, and the role of AI assistants in deep work tasks. Guillaume Marquis, CTO of VirtualBrain, specifically mentions using Qdrant due to its open-source nature, security aspects, and ability to handle terabytes of data with fast ingestion and semantic search.
  42. Talk with YouTube without paying a cent - Francesco Saverio Zuppichini | Vector Space Talks - QdrantThis post details the technical implementation of a local RAG system that ingests YouTube video subtitles for conversational querying. It showcases the use of YouTube DL for subtitle extraction, Hugging Face for generating embeddings, and Qdrant for storing and retrieving these embeddings locally. The author emphasizes a 'no framework' approach, preferring direct use of libraries and local deployment of Qdrant and Ollama for LLM inference. Key technical points include the preference for functional programming over classes for code organization, the strategic use of metadata as payload in Qdrant, and the practical aspects of setting up and interacting with a local Qdrant instance via its client. The post also briefly touches on organizing data into separate collections based on data types and the importance of proactive code understanding for scalability.
  43. Production-scale RAG for Real-Time News Distillation - Robert Caulk | Vector Space Talks - QdrantThis post highlights the use of Qdrant in a production-scale RAG system for real-time news distillation, emphasizing the challenges of context engineering and the need for robust, real-time indexing. It showcases how Qdrant enables efficient semantic similarity searches and extraction of important facts from a large volume of news articles, contrasting it with the limitations of traditional web search and all-in-one solutions. The discussion touches upon the importance of a clean vector space for semantic search and the benefits of Qdrant's modularity and performance for orchestrating such systems.
  44. Integrating Qdrant and LangChain for Advanced Vector Similarity Search - QdrantThis post details the integration of Qdrant with LangChain to build Retrieval Augmented Generation (RAG) applications, focusing on providing long-term memory for GenAI apps. It explains how LangChain simplifies RAG development by unifying interfaces to embedding models and vector stores like Qdrant. The post outlines common use cases such as chatbots, recommendation systems, and data analysis, and provides a tutorial for building a RAG chatbot using Qdrant, LangChain, and OpenAI. It also touches upon scaling Qdrant and LangChain for production environments.
  45. IrisAgent and Qdrant: Redefining Customer Support with AI - QdrantIrisAgent integrated Qdrant as their vector database of choice for their AI-powered customer support automation platform. They leveraged Qdrant's open-source nature, performance, ease of use (Go and Python SDKs), and maintainability (Rust) to augment their proprietary AI models for tasks like hallucination detection, similarity detection, and intent classification. Qdrant's integration enhanced their RAG use cases, leading to significant performance and quality gains. They are deploying Qdrant on Google Cloud.
  46. DSPy vs LangChain: A Comprehensive Framework Comparison - QdrantThis post provides a comparative analysis of LangChain and DSPy, two prominent frameworks for building AI applications. It details the features, performance, and use cases of each framework, focusing on their respective approaches to LLM integration, retrieval, and composition. LangChain is presented as a versatile orchestration framework with broad support for data sources and APIs, utilizing LCEL for chaining components. DSPy is highlighted for its ability to automate prompt optimization and tuning, eliminating manual prompt engineering. The post guides readers on selecting the appropriate framework based on project type, technical expertise, and community support, positioning Qdrant as a key vector store integration for both.
  47. The Bitter Lesson of Retrieval in Generative Language Model Workflows - Mikko Lehtimäki | Vector Space Talks - QdrantThis post discusses the 'bitter lesson' of retrieval in generative language model workflows, emphasizing the importance of data and compute over human-designed features. It highlights the development of Yokot AI, an LLM-based productivity booster that connects to enterprise data, and its retrieval-augmented generation (RAG) mechanism. The post details how Yokot AI manages diverse data inputs, scrapes the web, and leverages re-ranking to improve LLM output quality. It also touches upon the challenges of open-ended data inputs and languages in RAG systems, and the evolution of information retrieval methods.
  48. Vector Search Complexities: Insights from Projects in Image Search and RAG - Noé Achache | Vector Space Talks - QdrantThis post details insights from projects involving image search and Retrieval Augmented Generation (RAG). It highlights the efficacy of Dino V2 for image representation without fine-tuning, and discusses challenges in deduplicating real estate listings and document retrieval. The post also touches upon handling sensitive medical data and the need for new model types. Specific technical discussions include using Dino V2 for image feature extraction, comparing it to traditional methods like SIFT, and the benefits of asynchronous API calls for embedding generation.
  49. How to Superpower Your Semantic Search Using a Vector Database Vector Space Talks - QdrantThis post details Malt's transition to a retriever-ranker architecture for their freelancer matching application, leveraging Qdrant to significantly reduce search latency from 10 seconds to 1 second. The implementation involved adopting a multilingual transformer-based encoder model to generate embeddings and integrating Qdrant within a Kubernetes deployment managed by Argo CD. The core technical contribution is the successful application of Qdrant for semantic search in a large-scale marketplace, demonstrating performance and scaling improvements.
  50. Building LLM Powered Applications in Production - Hamza Farooq | Vector Space Talks - QdrantThis post is a transcript of a talk by Hamza Farooq, founder of Traversaal.ai, discussing building LLM-powered applications in production. It covers complex search, conversational search, personalization, challenges in LLM service downtime, and the importance of UX in GenAI products. While it mentions Qdrant as a tool for building these applications, it does not introduce new technical capabilities or significant technical depth about Qdrant's internal systems or specific engineering challenges. The focus is on the application layer and the broader ecosystem of LLM development.
  51. When music just doesn't match our vibe, can AI help? - Filip Makraduli | Vector Space Talks - QdrantThis post details the application of Qdrant in a music recommendation system. It describes using large language models to generate descriptive text for songs, encoding these descriptions into embeddings using sentence transformers, and then using Qdrant for cosine similarity search to find similar songs based on user-provided mood descriptions. The post highlights the use of Qdrant's vector similarity search capabilities for generating personalized recommendations.
  52. Chat with a codebase using Qdrant and N8N - QdrantThis post introduces the integration of Qdrant with n8n, a visual workflow automation tool, to facilitate the creation of agentic retrieval systems. It provides step-by-step instructions for building two n8n workflows: one for ingesting GitHub repositories into Qdrant using nodes like 'Qdrant Vector Store - Insert', 'GitHub Document Loader', 'Embeddings OpenAI', and 'Recursive Character Text Splitter', and another for creating a chat service with the ingested documents using 'Qdrant Vector Store - Retrieve', 'Retrieval Q&A Chain', 'Embeddings OpenAI', and 'OpenAI Chat Model'. This demonstrates a practical, low-code approach to building RAG-like applications.