Job opportunity -Gemspring - Sr MLOps Engineer - Remote

Client: Investment Company

Job Title : Sr MLOps Engineer

Location: 100% Remote

Job Summary:

Core Technical Skills

ML Pipeline & Model Lifecycle

• ML pipeline design and orchestration MLflow, Kubeflow, Azure ML Pipelines
• Model deployment across cloud and hybrid environments including model serialisation (ONNX, Pickle, TorchScript, SavedModel) and serving infrastructure design. Knowing how to package a model correctly for the target runtime is a baseline expectation, not a bonus
• Model scaling horizontal and vertical scaling strategies for inference workloads, autoscaling configurations, and load balancing for high-availability model serving across portfolio company environments
• Feature store management and experiment tracking for reproducibility
• LLM-specific operations: prompt versioning, RAG pipeline observability, evaluation frameworks, and fine-tuning pipeline management

Containerisation & Orchestration

• Docker hands-on experience containerising ML models and pipelines. Must be comfortable writing Dockerfiles, managing images, and designing container-based deployment workflows for model serving
• Kubernetes working knowledge of deploying and managing containerised workloads in Kubernetes environments. Does not need to be a Kubernetes administrator, but must understand deployments, services, resource limits, and horizontal pod autoscaling as they apply to ML model serving
• Familiarity with Kubernetes-native ML tooling (Kubeflow, KServe, Seldon) is a plus given the Azure-dominant environment across GemSpring's portfolio

Model Monitoring & Quality

• Data drift and concept drift detection identifying when model inputs or outputs have shifted enough to degrade performance
• Prediction quality monitoring and alerting systems that give portfolio company teams early warning before failures surface in production
• Model retraining pipeline design structured, automated pathways for updating models as conditions change

Data Architecture Foundations Critical Requirement

• Strong data architecture fundamentals Abhilash Vantaram flagged this explicitly: most AI engagements fail because of improper data architectures. This person must understand data modelling, pipeline design, and storage patterns well enough to assess an existing data environment and identify structural risks before they become deployment failures
• If this candidate does not carry strong data architecture depth, InfoVision recommends adding a dedicated Data Architect to the Tiger Team this risk must be covered by someone

Enterprise Integration

• Deep ERP-specific expertise is not required familiarity with enterprise integration patterns is understanding what connectors, MCPs, and cloud bridges already exist and leveraging them rather than rebuilding from scratch
• Ability to design MLOps infrastructure that integrates cleanly with existing ERP and CRM data flows across SAP, Salesforce, NetSuite, and equivalent mid-market platforms
• Scope clarity: once AI is in production across multiple portfolio companies simultaneously, the model lifecycle demands a dedicated owner. This role is not the infrastructure engineer that is the DevOps Engineer. Splitting attention between pipelines and model lifecycle management produces underperformance in both; the two functions must stay distinct

AI-Driven Development Practices

• AI-assisted SDLC fluency uses AI tooling as a matter of course across the full software delivery lifecycle code generation and review (GitHub Copilot, Cursor, Claude), AI-assisted test writing, automated documentation generation, AI-driven PR review, and AI-powered security scanning. These are table-stakes instruments in a 2026 delivery environment, not differentiators. A candidate who is still treating them as optional is already behind the pace this team operates at
• Prompting as an engineering discipline chain-of-thought, few-shot, system-level, role-based, and structured output prompting patterns with the ability to version, evaluate, and iterate on prompts as first-class engineering artifacts. Ad hoc prompting is not the standard; repeatable, testable, and documented prompt design is. Knows how to evaluate prompt performance systematically rather than informally
• Hands-on cloud-based AI workflow implementation direct, production-grade experience building and deploying AI workflows on at least one major hyperscaler Azure AI Services, AWS Bedrock, or Google Vertex AI. Understands managed inference endpoints, AI orchestration services, serverless AI functions, and AI gateway patterns from having built them, not from reading about them. Azure is the dominant environment across GemSpring's portfolio and must be a primary area of fluency; exposure to a second hyperscaler is a strong differentiator given the portfolio's mix of legacy and cloud-native companies
• Design for the team that follows the portfolio company's engineers will maintain this after the Tiger Team moves on. Every pipeline, monitoring setup, and deployment pattern must be operable by people who weren't part of the original build
• Risk-first orientation data drift and model degradation are silent failures. The ability to design monitoring systems that surface problems early before they become visible to the business is the core value this role provides
• Containerisation as standard practice Docker is not optional every model deployment should be containerised from the outset. This person must treat containerised, reproducible deployments as the baseline, not a nice-to-have. Kubernetes knowledge is required to operate and scale those deployments reliably in production
• Adaptability across maturity levels will work with companies ranging from those with mature data platforms to those with minimal data infrastructure. Must deliver meaningful MLOps capability in both contexts without requiring the company to reach an ideal state first

Experience Baseline

• 7+ years in ML engineering or MLOps roles with demonstrated production model management experience
• LLM deployment and observability experience strongly preferred
• Strong data architecture fundamentals verifiable, not just stated