Senior AI Ops / MLOps Engineer

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