Real-time Computer Vision at Scale

1 post · Apr 2024 – Jan 2026

What this trail covers

Qdrant enables real-time multimodal similarity search for fraud and abuse detection at Flipkart, processing high-dimensional embeddings from images and other data sources to reduce detection time from hours to minutes. This capability is being expanded for company-wide RAG systems, standardized Kubernetes deployments, and agentic AI workflows. VISUA leverages Qdrant for quality control and anomaly detection in computer vision data, achieving 40x faster query processing and a 10x scalability boost.

All Qdrant trails

Jan 2026Building real-time multimodal similarity search in Flipkart Trust & Safety with Qdrant - QdrantThis post details Flipkart's implementation of Qdrant for real-time multimodal similarity search in their Trust & Safety team. It highlights the limitations of their previous HBase+LSH batch-based system for high-dimensional embeddings and the selection of Qdrant due to its deployment flexibility, efficient HNSW indexing for simultaneous reads/writes, and support for high-dimensional data. The implementation resulted in a reduction of detection time from 9 hours to under 1 minute, with simplified integration via the Java gRPC SDK and Prometheus metrics. Future plans include expanding RAG use cases, standardizing Kubernetes deployments, and exploring agentic AI integrations.
Nov 2025How Dragonfruit AI scaled real-time computer vision with Qdrant - QdrantThis post details how Dragonfruit AI uses Qdrant for real-time computer vision at scale. It highlights the challenges of processing high FPS from numerous cameras on edge devices and transmitting only compact inference data to the cloud. Key Qdrant features leveraged include per-collection configurability for diverse workloads (ingestion-heavy vs. read-intensive), efficient float16 vector formats for memory and cache optimization, and its open-source nature for ecosystem integration. The solution architecture involves Qdrant serving distinct retrieval patterns for multi-camera person re-identification, self-checkout product verification, and full-frame semantic search. The results include faster agent development, lower total cost of ownership, and the ability to handle real-time processing of many cameras per site at high FPS on edge devices with optimized data transmission.
May 2024Visua and Qdrant: Vector Search in Computer Vision - QdrantThis post details VISUA's use of Qdrant for quality control and anomaly detection in computer vision data. They achieved 40x faster query processing and a 10x scalability boost by leveraging Qdrant's hybrid query capabilities (vector similarity + payload filtering) and performance superiority over alternatives like Weaviate and Pinecone. Qdrant's support for multiple vectors per record and quantization were key implementation details.
Advertisement