At Navan, we aren't building a single, generic chatbot. We are building a Composable AI Microservice Architecture, a swarm of hundreds of hyper-specialized AI services, each meticulously "programmed" to solve small, focused tasks with high precision. This fleet powers Ava, our AI support engine, and a suite of cutting-edge generative tools for travel and expense management.
As a Senior AI Ops / MLOps Engineer, you are the architect of the platform that makes this scale possible. You will move beyond traditional MLOps to manage a "factory" of Language Models. Your challenge is one of orchestration and standardization, ensuring that every service in the swarm meets a rigorous bar for quality, reliability, and cost-efficiency.
What You'll Do
- Orchestrate the AI Fleet: Build and own the runtime environment for 100+ specialized AI services. Manage model routing, context versioning, and standardized memory/history stores.
- High-Density Inference Optimization: Design and implement SageMaker Multi-Model Endpoints (MME) and Inference Components to serve multiple tuned SLMs per GPU, maximizing hardware utilization while minimizing latency.
- Deterministic Service Excellence: Treat reliability as a layered engineering problem. Build deterministic "shells" around probabilistic LM outputs, prioritizing data-layer validation and strict serialization.
- Automated Evaluation & Observability: Implement "LLM-as-a-judge" patterns and automated benchmarking to detect semantic drift and hallucinations across the fleet before they impact the user.
- Standardize the Workflow: Obsess over building reusable patterns and Terraform-based infrastructure that eliminate "snowflake" configurations, allowing us to deploy new specialized AI tasks in minutes.
- Agency Strategy: Partner with AI Researchers to find the "Goldilocks zone" for agentic autonomy—balancing the flexibility of LLM tool-use with the precision required for production stability.
What We're Looking For
- Experience: 5+ years in SRE, Platform Engineering, or MLOps, with at least 2 years focused on deploying LLMs/SLMs in production environments.
- SageMaker Mastery: Deep hands-on expertise with AWS SageMaker, specifically configuring Multi-Model Endpoints (MME), Inference Components, and GPU-backed instances (G5/P4).
- SLM Expertise: Proven experience with Small Language Models (e.g., Mistral, Llama 3, Phi) and parameter-efficient fine-tuning (PEFT) deployment strategies like LoRA/QLoRA.
- Technical Stack: * Languages: Strong proficiency in Python and Terraform.
- Orchestration: Experience with Docker, Kubernetes (EKS), or AWS ECS/Fargate.
- Data: Familiarity with Snowflake and Vector Databases.
- The "AI Ops" Mindset: You understand that AI at scale is a statistical challenge. You are comfortable debugging issues at the data/serialization layer rather than defaulting to prompt tweaks.
- CI/CD & Automation: Experience building robust pipelines (Jenkins, GitHub Actions) for non-deterministic software, including automated "eval" stages.
- Education: BS or MS in Computer Science, Engineering, Mathematics, or a related technical field.
Must have
- Python, Terraform, Sagemaker