Teams at Hyphen Connect Limited

Optimize and compress large language and vision models for on-device inference. Build distillation and hardware-specific compilation pipelines, benchmark performance across NPU/GPU architectures, and deploy models to edge environments.

Design and integrate vision encoders and audio-native models into agent reasoning loops, optimize streaming latency for voice-to-voice interactions, and architect multimodal RAG systems to retrieve insights from videos and PDFs while ensuring efficient cross-modal alignment.

Design, orchestrate, and optimize large-scale LLM pre-training runs across 1,000+ GPUs using PyTorch/DeepSpeed/Megatron-LM. Improve networking (InfiniBand/RDMA) and memory management, implement 3D parallelism, and automate checkpointing and failure recovery for month-long training jobs on SLURM or Kubernetes GPU clusters.