November 2025 Summaries

Cosmos, a visual search app aimed at creative professionals and everyday users, has enhanced its search capabilities by integrating Qdrant Cloud to deliver precise and engaging visual discovery experiences. The platform focuses on high-quality, human-made content, providing a meditative and ad-free space for idea curation, contrasting with typical algorithm-driven feeds. To support its community of creatives, Cosmos implemented Qdrant Cloud to manage text, color, and hybrid visual searches with sub-second latency, addressing scalability and performance challenges previously encountered with Postgres. Qdrant facilitates multimodal retrieval by storing various embeddings like CLIP, CNN, and color vectors in a single collection, allowing Cosmos to optimize relevance, engagement, and aesthetics in its search results. The app's exact color search system uses CIELAB vectors to ensure precise color matching, vital for the platform's aesthetic-driven approach. With these advancements, Cosmos has achieved faster processing and predictable operations, scaling effectively with user growth, while preparing for future expansions in AI capabilities and app relaunches.

Qdrant version 1.16 introduces several significant updates aimed at enhancing its multitenancy, search efficiency, and user experience. The new Tiered Multitenancy feature allows for a more flexible approach by combining small and large tenants within a single collection, promoting growing tenants to dedicated shards, which balances resource allocation and performance. The introduction of ACORN improves the quality of filtered vector searches, particularly when multiple filters with low selectivity are applied, by examining neighbors of neighbors in the HNSW graph, albeit with some performance overhead. Version 1.16 also debuts inline storage, which optimizes disk-based vector searches by embedding quantized vector data directly within HNSW nodes, thereby reducing random access reads and enhancing search performance. Enhancements to full-text search include the addition of the text_any condition for simpler matching of multiple terms and ASCII folding for better handling of multilingual texts. Furthermore, a new conditional update API helps prevent unintended data overwrites during concurrent updates, and a redesigned Web UI improves user interaction with the platform. These updates collectively aim to boost Qdrant's performance and usability, catering to the growing demands of vector search applications.

Xaver has addressed the growing demand for personalized financial advice by developing an AI-driven knowledge engine built on the Qdrant platform, designed to deliver real-time, compliant guidance across various digital channels. This system includes a two-tier retrieval architecture, comprising a condensed knowledge base for quick responses and a full knowledge base for more complex inquiries, which minimizes latency and ensures high-quality interaction. By leveraging Qdrant's capabilities, Xaver has successfully scaled its advisory services without compromising on performance or compliance, allowing financial institutions to modernize customer interactions with a digital-first approach while maintaining the trust and regulatory standards of traditional consultation methods.

Dragonfruit AI has effectively scaled its real-time computer vision solutions by integrating the open-source vector search engine, Qdrant, into its platform, which transforms IP camera feeds into actionable insights for various enterprise applications such as security and operations. Utilizing a Split AI approach, Dragonfruit conducts real-time inference on-premises for efficiency and uses cloud services for data aggregation and search, thereby minimizing total costs and latency while maintaining existing customer infrastructure. Qdrant's per-collection configurability and efficient numeric formats have enabled Dragonfruit to manage diverse workloads efficiently, achieving real-time performance and cost-effective storage. This integration has allowed the rapid development of new AI agents for loss prevention and safety, broadening their market reach across multiple sectors. By leveraging Qdrant, Dragonfruit has optimized memory usage with float16 embeddings and enhanced throughput without sacrificing retrieval accuracy, demonstrating the operational synergy between tailored vector search and AI-powered analytics at an enterprise scale.