About the role
Responsibilities:
- Architect and scale a next-generation multimodal inference platform powering multiple production AI pipelines.
- Building GPU-resident multimodal pipelines for image, audio, and video inference.
- Optimizing PyTorch models with TensorRT and Torch-TensorRT.
- Designing dynamic batching and low-latency inference systems.
- Building GPU-native preprocessing pipelines using NVIDIA DALI and Kornia.
- Supporting inference across heterogeneous GPU fleets (H100 A100 A10G, RTX 4090 etc. ).
- Improving observability, throughput, reliability, and deployment automation.
Requirements:
- Strong Python and PyTorch skills.
- Experience deploying ML models in production.
- Understanding of GPU inference optimization and CUDA fundamentals.
- Familiarity with Docker and Linux.
- Strong debugging and problem-solving skills.
- Experience with NVIDIA Triton, TensorRT, or DALI.
- Familiarity with CUDA profiling and performance optimization.
- Experience with distributed inference systems.
- Knowledge of audio/video ML pipelines.
- Experience with ONNX, AWS GPU infrastructure, and model quantization techniques (FP16/INT8).