H
HYPERCONNECT Tech Stack
Video platform with real-time ML recommendation and live-streaming infrastructure
HYPERCONNECT operates a global video messenger (Azar) backed by a production ML stack spanning Kafka, Flink, and GPU inference (TensorFlow, PyTorch, NVIDIA Triton). The adopting list—DeepSpeed, TensorRT, KEDA—shows aggressive optimization of model serving costs and cluster utilization, directly addressing their stated pain around GPU idle time. Active projects (real-time session recommendations, multimodal LLM labeling, continuous learning pipelines) reveal a data team scaling inference velocity to match live-traffic demand.
Tech Stack 52 technologies
Core StackKotlin Redis Kafka Braze Apache Flink Java Go Python TypeScript BigQuery Apache Airflow Tableau ArgoCD TensorFlow PyTorch Prometheus Grafana Ansible Terraform Spring Boot JPA SQL KSQL Looker Studio NVIDIA Triton Argo Workflows Slurm InfiniBand Lustre NFS+17 more
AdoptingDeepSpeed NVIDIA TensorRT ONNX Runtime Triton Inference Server KEDA
What HYPERCONNECT Is Building
◆Challenges
- Real-time event data processing
- Traffic spikes
- Performance bottlenecks
- High availability
- Optimizing model serving cost
- Reducing gpu cluster idle time
- Reducing ml lifecycle inefficiencies
- Hiring process inefficiency
- Talent pool scarcity
- Redefine business problems as ml problems
▲Active Projects
- Recommendation model scaling
- Real-time inference data pipeline
- Cupid: real-time session-based recommendation system
- Live streaming recommendation system
- Multimodal llm labeling research
- Continuous learning pipeline
- Crm strategy implementation
- A/b testing design and verification
- Airflow-based automation pipeline
- Dashboard design with tableau/looker studio
Hiring Activity
Minimal15 roles · 1 in 30d
Department
Seniority
Company intelligence
