Client: Global newswire and media organization.Their content reaches more than half the global population daily. The tech org is modern and investing heavily in ML infrastructure to support large-scale media processing across text, image, and video.
Role Description
This is a senior hands-on ML engineering job focused on building and optimizing inference systems that run in production at scale. You'd be working across text, image, and video pipelines - processing millions of media assets to power news intelligence products. Think DistilBERT for NER, SBERT for embeddings, TransNetV2 for video shot detection, and external multimodal APIs for captioning.
This is not an MLOps or platform role. They need someone who can profile a transformer, rewrite its serving path for a 2-3x latency improvement, tune an HNSW index, and make smart infrastructure decisions on SageMaker instance selection to hit p95 targets at the lowest cost. If your background is primarily Terraform, Kubernetes admin, or CI/CD pipelines - this isn't the right fit.
You'll partner closely with MLOps, platform engineering, data scientists, and product teams - but ownership of model performance, inference logic, and pipeline efficiency lives here.
Required Skills
5+ years building production ML inference systems
Python - core to everything in this role
PyTorch (TorchScript, ONNX, FastAPI/TorchServe) and TensorFlow (SavedModel, tf.data, XLA, TFLite) - both required
Deep hands-on experience with transformer-based models (BERT family - DistilBERT, SBERT, etc.) in production
Inference optimization at scale - quantization, distillation, compilation, kernel/profile-level performance work
AWS infrastructure - EC2, Batch, Lambda, SageMaker across different media workload types