S
Sciforium Tech Stack
Byte-native AI models and serverless LLM serving infrastructure
Sciforium builds foundation models and LLM serving infrastructure, with a deep systems focus: the stack spans PyTorch, JAX, CUDA, ROCm, vLLM, and distributed training frameworks (Ray, Kubernetes). The research-heavy org (5 engineers, 1 researcher, 1 data scientist) is tackling inference latency, distributed training optimization, and GPU kernel development — pain points that reflect a build-from-metal approach rather than API-wrapping. Backing from AMD signals hardware partnership potential.
Tech Stack 32 technologies
Core StackPyTorch Python C++ Kubernetes TypeScript Stripe FastAPI Linux JAX TPU Flax XLA CUDA Ray ROCm NVIDIA Nsight vLLM AMD WebSockets Server-Sent Events Ubuntu CentOS Red Hat Enterprise Linux cuDNN NCCL NFS GPFS Lustre LDAP FreeIPA+2 more
AdoptingStripe
What Sciforium Is Building
◆Challenges
- Scaling ai infrastructure
- Scaling full stack for frontier ai models
- Real-time inference performance
- Optimizing distributed training performance
- Debugging hardware-software interactions
- Scaling large-scale ai training workloads
- Performance optimization at scale
- Optimizing inference latency
- Building large etl pipelines for internet-scale data
- Maintaining gpu compute cluster health
▲Active Projects
- Post-training & alignment data
- Large language models research
- Generative media research
- Scalable training systems research
- Model serving platform
- Distributed inference systems
- Gpu kernel development
- Distributed training system optimization
- Ml software stack maintenance
- Llm training and deployment
Hiring Activity
Minimal10 roles · 0 in 30d
Department
Seniority
Company intelligence
