Real-Time Analytics
Sub-second dashboards and data products on petabytes of data at any concurrency
Data Warehousing
Sub-second analytics on open lakehouse formats with no vendor lock-in
Observability in the AI Era
The most cost-effective alternative to Elasticsearch observability
Context Engineering
Hybrid search and fresh context for RAG, agents, and LLMs
Financial Services
Real-time analytics and fraud detection for banking, capital markets, and on-chain data
Ad & Media tech
Advertiser-facing reporting, content search, and game telemetry on one engine
Automobile & Transportation
Fleet telemetry, real-time recommendations, and multimodal autonomous driving search
A connected car streams telemetry by the second, riders generate millions of live events, and autonomous driving programs produce petabytes of sensor footage. VeloDB, the real-time analytics and search database from the creators of Apache Doris, serves analytics, point lookups, and multimodal search behind these workloads in one engine.
Submit your inquiries here, and our team will get back to you
Tuhu Car (HKEX: 9690) serves 126 million car owners through an online-to-offline model, on a stack that ran Hive, HBase, MySQL, and Trino side by side with T+1 freshness and metrics that needed reconciliation. Consolidating on one OLAP engine took audience segmentation from more than 60 seconds to under 3, and data updates to minute-level.
BYD hit latency and concurrency limits on its mobile sales dashboard, battery cell performance traceability, and queries over vehicle-connected data, with high storage costs for connected vehicle analytics. VeloDB now serves its real-time and ad hoc analytics.
Xanh SM, Vietnam's leading EV ride-hailing platform, wanted destination recommendations that react to what is happening in the city right now. One storage layer with isolated compute groups runs both the ETL analytics and the live serving API.
Powering real-time data for 10,000+ teams
Vehicle data arrives as images, point clouds, video, labels, logs, and structured records. When each type lives in its own engine, teams copy the same data between systems, pipelines multiply, and the same metric returns different numbers depending on where you ask.
Sales dashboards, battery traceability, and vehicle-connected queries only help when the data is current. Batch pipelines deliver yesterday's answer, and fast ingestion alone just moves the bottleneck to the query engine.
Millions of app users hit profile lookups and recommendations while engineers mine billions of driving events for model training. Both workloads expect sub-second answers, and both spike without warning.
Scalar filters, JSON queries, full-text search, and vector retrieval run in a single SQL engine, so one query can combine scene attributes, sensor metadata, and embedding similarity. Apache Doris 4.0 added BM25 relevance scoring, and 4.1 adds a search() function with Elasticsearch-compatible DSL plus IVF vector indexes that reach trillion-vector scale. ByteDance serves 1 billion 768-dimension vectors from roughly 500 GB of memory on a single server, at about 400 ms p95 for hybrid retrieval.
Merge-on-write updates keep tables queryable the moment data lands. Native CDC ingests from operational databases and Kafka delivers streaming events, so freshness holds at minute level or better while queries stay fast. Xanh SM writes 175,000 records per second in under 1 second, and Tuhu moved data updates from T+1 to minute-level processing.
Point queries reach 30,000 QPS per node, complex joins run distributed across the cluster, and dashboards stay responsive through peaks, so customer-facing lookups and engineering analysis share one engine. A leading autonomous driving company holds sub-second responses at concurrent loads approaching 1,000 QPS across nearly 1 trillion records over a seven-day window, while its algorithm engineers need visibility into a million frames per second.
Tuhu serves 126 million car owners on one OLAP foundation. Segmentation queries improved 20x, profile lookups stabilized around 10 ms, and BI P90 latency dropped from 52 seconds to 7 seconds. Read here
Tuhu CarOn Vietnam's National Day in September 2025, Xanh SM surfaced the parade rehearsal venues with no manual input, and had already surfaced a concert venue as fans were heading there. Average query latency holds around 76 ms. Read here
Xanh SMA car manufacturer streams CAN data from 4 million connected cars, sustaining millions of TPS and dozens of terabytes a day, and joins around 1 billion daily diagnostic trouble codes against a live MySQL configuration table in one query. Read here
Car manufacturerA leading autonomous driving company replaced a batch warehouse, a dedicated vector database, and Elasticsearch with one engine, cutting query times from minutes to seconds across hundreds of millions of clips and tens of billions of training frames. Read here
Autonomous drivingAn EV charging network runs its log analytics on VeloDB, serving search and analytical queries from one engine. Story publishing soon
EV chargingVeloDB is a real-time analytics and search database from the creators of Apache Doris. One engine handles point lookups, complex joins, high-concurrency aggregations, and full-text and vector search on data that changes by the second.
Telemetry arrives continuously and analysis has to keep up. A car manufacturer streams CAN bus and sensor data from 4 million connected cars into Doris, sustaining millions of TPS and dozens of terabytes a day, with monitoring and reporting served from the same layer. Modelling the tables on vehicle identification number and timestamp replaced a Spark SQL flat-table transformation that previously ran once a day.
Training data retrieval mixes scene attributes, labels, logs, and embeddings. VeloDB runs scalar, JSON, full-text, and vector search in one SQL engine. A leading autonomous driving company uses it for stable retrieval across nearly 1 trillion records over a seven-day window at concurrent loads approaching 1,000 QPS, having replaced a batch warehouse, a separate vector database, and Elasticsearch.
Yes. Xanh SM ingests rider click events at spikes of up to 20,000 messages per second, writes 175,000 records per second in under 1 second, and serves personalized destination recommendations with average query latency around 76 ms, running the ETL jobs and the serving API on separate compute groups over one storage layer.
Inverted indexes serve full-text queries next to SQL aggregations, so log search and metric analysis run on one copy of the data. Apache Doris 4.0 added BM25 relevance scoring, and 4.1 adds a search() function with Elasticsearch-compatible DSL. An EV charging network runs its log analytics this way, pairing real-time ingestion with high-performance search.
