Senior MLOps Engineer

About the Company

At deRecruiters, we’re searching for a Senior MLOps Engineer for a travel tech scaleup that is a pioneer in airline inventory and post-booking revenue management technology. The company helps airlines move from static to dynamic post-booking inventory, so they can increase conversion, improve revenue optimization and give passengers more flexibility. Its platform connects existing airline systems with new inventory logic. This enables upselling, seat upgrades and flight swaps without disrupting core infrastructure. The company works with leading low-cost and regional carriers across LATAM and Asia.

Mission

The Senior MLOps Engineer will turn ML into a measurable, industrialized capability across the company. They will own the ML framework that takes every model from experiment to production. They will also build the measurement that links model performance to business performance. Finally, they will lay the foundation for agentic data science, where agents run experiments under human supervision.

Primary Focus

The role focuses on four areas: the ML framework and pipeline on GCP, ML performance measurement, production ML operations, and the DS Expert agent. The engineer will work hand in hand with the company’s data scientists, each of whom specializes in one ML model across all airline tenants.

Key Responsibilities

ML Framework & Pipeline

• Take over ownership of the ML framework and scale it from a validated sandbox cycle to the production path for every model: full-flight / spill prediction, expected resale value, elasticity, and the new swap-quality and spill-seat models.

• Run and evolve one pipeline from config to production:

  • Immutable data snapshots as the training input.

  • Training on Vertex AI Pipelines, with every run logged.

  • Evaluation through the serving code, so offline and production results match.

  • A declarative promotion gate against the champion, with confidence intervals per days-to-departure bucket.

  • Versioned, aliased model registration.

  • Releases via canary, then shadow, then promotion, with a human approval step.

  • Weekly drift and decision monitoring.

• Move from one code branch per tenant to one repo and one image, with the tenant as configuration.

• Guarantee full traceability (prediction → model version → run → snapshot → commit) and rollback by traffic flip within a minute.

• Maintain the in-house feature store and model configuration in BigQuery, in place of managed feature-store services.

ML Performance Measurement, Built Agentically

• Use AI coding agents to build the measurement that attributes business performance to the predictive performance of the company’s ML models. This covers each step of the operational funnel: seats freed, offers made, conversion, resale and economics.

• Compare live models, candidate models and a no-ML baseline on the same flights, so every model change has an expected business impact before it ships.

• Report ML-attributed uplift per airline every month, and publish results as governed datasets for people and agents.

Production ML Operations

• Serve models in production from the pipeline, rolling them out tenant by tenant.

• Implement monitoring for accuracy, data and concept drift, latency, throughput and cost, with alerting, rollback and retraining triggers.

• Keep ML workloads scalable, secure and cost-efficient on GCP.

• Define and own the batch prediction contract as the product moves from per-PNR real-time scoring to a daily batch pipeline, together with the platform and engineering teams.

Agentic Data Science: The DS Expert

• Build the DS Expert agent on top of the pipeline and the performance measurement. The agent will automate experiments and feature engineering so data scientists iterate faster.

• Package ML artifacts, configs and experiment know-how as governed skills in the company’s knowledge layer, so agents and people work from the same context.

• Keep humans in the loop where it matters: launching experiments and approving promotions.

Collaboration

• Partner with data scientists to co-design features, training pipelines and evaluation strategies, so they can focus on modeling, not infrastructure.

• Work with Data Engineering on data snapshots, gold tables and aligned release cycles.

• Work with the Performance team so ML attribution matches how product performance is reported.

• Expose models to agentic AI engineers through clean, stable APIs for real-time and batch use.

Core Tools

• Google Cloud: Vertex AI Pipelines and Model Registry, BigQuery, Cloud Storage

• Python (ML libraries, packaging, testing) and SQL (BigQuery)

• CI/CD and containers for ML releases

• Claude Code and agentic workflows (skills, MCP)

• Jira for change management

Ideal Candidate

• 5+ years in ML engineering or MLOps, with models they have taken to production and operated.

• Hands-on with Vertex AI (or an equivalent ML platform) and BigQuery.

• Strong CI/CD for ML: automated testing of data, features and models, versioned deployments, controlled rollouts.

• Experience with champion/challenger, shadow deployments and statistically sound promotion rules.

• Able to build counterfactual measurement that links model quality to business outcomes.

• Strong Python engineering and SQL; writes reproducible, auditable pipelines.

• AI-native: uses coding agents daily to multiply output, and has built or wants to build agents that automate engineering and data science work.

• Clear ownership and communication with data scientists, engineers and business stakeholders.

Nice to Have

• Airline revenue management, pricing or demand forecasting experience.

• Experience building LLM agents end to end: context and skills design, tools, guardrails and evaluation.

• Experience migrating ML workloads from real-time microservices to batch pipelines, or from AWS to GCP.