Safety reads and high-volume writes fought for the same nodes. Milvus disaggregates them: 10B+ vectors, latency steady under load.
One of the world's largest short-video and live-streaming platforms, with 400 million daily active users and a sprawling AI stack spanning content safety, video deduplication, semantic search, and generative features, all requiring retrieval of unstructured content by meaning rather than keyword.
Their Elasticsearch-based vector layer couldn't separate reads from writes, causing query latency spikes in safety and risk systems exactly when...
“Kwai runs one of the largest Milvus deployments in our community, and its team has deep expertise in running Milvus at large scale. We're proud to support Kwai and excited to keep building together.”
Short-video sharing and live-streaming social platform.
Vector database platform for building and scaling AI applications.
Kwai's Vector search platform is part of this use case:
Related implementations across industries and use cases
Elasticsearch kept every vector in RAM—cost-prohibitive at scale. One shared Milvus platform now runs 100+ workloads on 10B+ vectors.
A phone launch, a breaking story, a Vietnamese phrase — keyword search came up short. Semantic retrieval now covers 92% of AI Search.
Daily updates delayed new content. A unified database now runs twice-daily refreshes with zero downtime.
Elasticsearch kept every vector in RAM—cost-prohibitive at scale. One shared Milvus platform now runs 100+ workloads on 10B+ vectors.
A phone launch, a breaking story, a Vietnamese phrase — keyword search came up short. Semantic retrieval now covers 92% of AI Search.
Complex headcount reporting meant tedious manual data slicing. Now, HR uses AI prompts to instantly generate CFO-ready talent insights.
Assessing crash damage bottlenecked repairs for 30 minutes. Now, AI drafts parts lists in 5 minutes for human review, halving repair time.
Tournaments running simultaneously meant an hour of manual checks each. AI agents now run them in minutes, freeing the team to be proactive.
Large AI training jobs meant fighting for preemptible slots or leaving campus. Marlowe gave any lab guaranteed multi-node access on demand.
Safety reads and high-volume writes fought for the same nodes. Milvus disaggregates them: 10B+ vectors, latency steady under load.
One of the world's largest short-video and live-streaming platforms, with 400 million daily active users and a sprawling AI stack spanning content safety, video deduplication, semantic search, and generative features, all requiring retrieval of unstructured content by meaning rather than keyword.
Their Elasticsearch-based vector layer couldn't separate reads from writes, causing query latency spikes in safety and risk systems exactly when...
“Kwai runs one of the largest Milvus deployments in our community, and its team has deep expertise in running Milvus at large scale. We're proud to support Kwai and excited to keep building together.”
Short-video sharing and live-streaming social platform.
Vector database platform for building and scaling AI applications.
Kwai's Vector search platform is part of this use case:
Related implementations across industries and use cases
Elasticsearch kept every vector in RAM—cost-prohibitive at scale. One shared Milvus platform now runs 100+ workloads on 10B+ vectors.
A phone launch, a breaking story, a Vietnamese phrase — keyword search came up short. Semantic retrieval now covers 92% of AI Search.
Daily updates delayed new content. A unified database now runs twice-daily refreshes with zero downtime.
Elasticsearch kept every vector in RAM—cost-prohibitive at scale. One shared Milvus platform now runs 100+ workloads on 10B+ vectors.
A phone launch, a breaking story, a Vietnamese phrase — keyword search came up short. Semantic retrieval now covers 92% of AI Search.
Complex headcount reporting meant tedious manual data slicing. Now, HR uses AI prompts to instantly generate CFO-ready talent insights.
Assessing crash damage bottlenecked repairs for 30 minutes. Now, AI drafts parts lists in 5 minutes for human review, halving repair time.
Tournaments running simultaneously meant an hour of manual checks each. AI agents now run them in minutes, freeing the team to be proactive.
Large AI training jobs meant fighting for preemptible slots or leaving campus. Marlowe gave any lab guaranteed multi-node access on demand.