Senior MLOps Engineer

Unico Connect is an AI-first product engineering agency building custom software, mobile apps, and AI solutions for clients. We are hiring a Senior MLOps Engineer to own the infrastructure and operational layer of the AI systems we ship: classical ML, LLM products, and agentic workflows.

You will define MLOps architecture for client engagements, stand up deployment, security, monitoring, and evaluation infrastructure, and ship production-grade AI that passes enterprise security review. You will also act as a technical voice in client conversations and translate requirements into defensible architecture decisions with clear reasoning on cost, scale, security, and maintainability.

Responsibilities

• AI infrastructure and architecture: Define MLOps architecture on AWS (primary), covering distributed training (DeepSpeed, FSDP, Ray), LLM gateways (LiteLLM, Portkey), model optimisation (quantisation, TensorRT-LLM, ONNX Runtime), and GPU cluster management.
• Training pipelines and data: Build reproducible pipelines with versioned data (DVC, LakeFS, Delta Lake), data validation as code, feature stores, experiment tracking (MLflow, W&B), and hyperparameter optimisation (Ray Tune, Optuna).
• Model deployment and serving: Package and deploy classical ML, LLMs, and multi-modal models via Triton, vLLM, TGI, KServe, BentoML, or TorchServe. Own inference scaling, latency, cost per inference, and failover.
• GenAI and agentic infrastructure: Build RAG pipelines, vector databases (Pinecone, Weaviate, Qdrant, pgvector, OpenSearch), embedding workflows, prompt versioning, agent infrastructure (LangGraph, CrewAI, AutoGen, MCP), and fine-tuning (LoRA, QLoRA, PEFT).
• AI security and compliance: Build to OWASP LLM Top ), Google SAIF, MITRE ATLAS, and NIST AI RMF. Implement guardrails, prompt-injection defenses, ML-BOM tracking, artifact signing, PII detection, network isolation, IAM, and audit logging. Support DPDP Act evidence.
• Testing and evaluation: Build the evaluation stack (Langfuse, Promptfoo, PyRIT, Giskard, TruLens). Gate deployments on golden datasets, regression tests, and bias/fairness checks (Fairlearn, AIF360). Run shadow, canary, and A/B deployments.
• Monitoring, AI SRE, and FinOps: Define SLIs and SLOs (p95 latency, refusal rate, hallucination rate, drift) with error budgets. Own runbooks, retraining triggers, rollback, and on-call. Establish FinOps discipline on cost per inference and token spend per tenant and feature.
• CI/CD for ML: Design pipelines across data, model, application, and evaluation layers with reproducibility, traceability, and security scanning built in.
• Responsible AI and governance: Produce model cards and datasheets. Implement explainability (SHAP, LIME, Captum). Support human-in-the-loop review and compliance evidence for different frameworks.
• Architecture, mentorship, and client engagement: Evaluate managed services versus custom builds with cost, scale, and security reasoning. Contribute to proposals, scoping, and client security reviews. Mentor mid-level and junior engineers through code and design reviews.

Requirements

• 5 years of hands-on MLOps experience, with at least one LLM or RAG application shipped to production for an enterprise client.
• AWS at depth: SageMaker (Studio, Pipelines, Endpoints, Feature Store, Model Registry), Bedrock, ECS or EKS, S3, Lambda, Step Functions, CloudWatch, IAM, KMS. Equivalent Azure or GCP considered with willingness to work in AWS.
• Strong Python with comfort reading and debugging ML code. Working knowledge of PyTorch or TensorFlow at infrastructure level (packaging, optimisation, quantisation).
• Docker, Kubernetes, Terraform, and one CI/CD platform (GitHub Actions, Jenkins, GitLab CI, or Azure DevOps).
• MLOps tooling across experiment tracking (MLflow, W&B, Neptune, SageMaker Model Registry), orchestration (Kubeflow, Airflow, SageMaker Pipelines, Metaflow), serving (Triton, vLLM, TGI, KServe, BentoML, TorchServe), and monitoring (Evidently, WhyLabs, Arize, Langfuse).
• GenAI stack: hands-on with at least two of LangChain or LlamaIndex, a vector database, an LLM evaluation framework, and guardrails libraries. Practical exposure to RAG, embedding workflows, and prompt versioning.
• Working knowledge of OWASP LLM Top 10, Google SAIF, or NIST AI RMF, with hands-on experience using at least one guardrail library and a PII-detection tool.
• Clear written and verbal communication. Able to explain architectural trade-offs to founders and product owners in their language.