AI AND MACHINE LEARNING ENGINEER

**Must Have:**

• 2+ years of experience in MLOps, DevOps, or backend engineering for AI workloads
• Proficiency in DeepStream 7.x including pipelines, Gstplugins, nvdsanalytics, nvstreammux
• Strong understanding of containerization (specifically Docker) and GPU scheduling
• Proven track record in optimizing latency/throughput on NVIDIA GPUs using technologies like TensorRT, mixed precision, and CUDA toolkit
• Hands-on experience in deploying YOLO or similar Convolutional Neural Networks in production environments
• Familiarity with self-hosting and serving LLMs such as vLLM, TensorRTLLM, or equivalent, along with quantization, pruning, and distillation
• Proficiency in Python and bash scripting, with confidence in CI/CD scripting

**Nice to have:**

• Exposure to cloud GPUs (AWS/GCP/Azure)
• Experience with edge devices like Jetson, Xavier, Orin
• Performance profiling using tools like Nsight Systems/DCGM
• Knowledge of Triton Inference Server internals
• Familiarity with distributed training methods like PyTorch DDP, DeepSpeed
• Basic skills in frontend/REST gRPC API design

In this role, you will:

• Build and automate inference pipelines
• Design, containerize, and deploy Computer Vision models (such as YOLO v8/v11, custom CNNs) with DeepStream 7.x, focusing on optimizing for lowest latency and highest throughput on NVIDIA GPUs
• Migrate existing Triton workloads to DeepStream with minimal downtime
• Serve and optimize large language models
• Self-host Llama 3.2, Llama 4, and future LLM/VLMs on the cluster utilizing best practice quantization, pruning, and distillation techniques
• Develop fast, reliable APIs and monitoring for downstream teams
• Implement continuous delivery and observability practices, automating build/test/release steps and setting up health metrics, logs, and alerts to ensure model stability in production
• Efficiently allocate GPU resources across CV and LLM services

- Provide model lifecycle support by assisting data scientists with occasional finetuning or retraining runs and packaging models for production.,