We are seeking an AI Platform Engineer to build and scale the infrastructure that powers our production AI services. You will take cutting-edge models, ranging from speech recognition (ASR) to large language models (LLMs)-and deploy them into highly available, developer-friendly APIs.
You will be responsible for creating the bridge between the R&D team, who train models, and the applications that consume them. This means developing robust APIs, deploying and optimising models on Triton Inference Server (or similar frameworks), and ensuring real-time, scalable inference.
Responsibilities
API Development
* Design, build, and maintain production-ready APIs for speech, language, and other AI models.
* Provide SDKs and documentation to enable easy developer adoption.
Model Deployment
* Deploy models (ASR, LLM, and others) using Triton Inference Server or similar systems.
* Optimise inference pipelines for low-latency, high-throughput workloads.
Scalability & Reliability
* Architect infrastructure for handling large-scale, concurrent inference requests.
* Implement monitoring, logging, and auto-scaling for deployed services.
Collaboration
* Work with research teams to productionize new models.
* Partner with application teams to deliver AI functionality seamlessly through APIs.
DevOps & Infrastructure
* Automate CI/CD pipelines for models and APIs.
* Manage GPU-based infrastructure in cloud or hybrid environments.
Requirements
Core Skills
* Strong programming experience in Python (FastAPI, Flask) and/or for API services.
* Hands-on experience with model deployment using Triton Inference Server, TorchServe, or similar.
* Familiarity with both ASR frameworks and LLM frameworks (Hugging Face Transformers, TensorRT-LLM, vLLM, etc.).
Infrastructure
* Experience with Docker, Kubernetes, and managing GPU-accelerated workloads.
* Deep knowledge of real-time inference systems (REST, gRPC, WebSockets, streaming).
* Cloud experience (AWS, GCP, Azure).
Bonus
* Experience with model optimisation (quantisation, distillation, TensorRT, ONNX).
* Exposure to MLOps tools for deployment and monitoring