Senior Applied ML Engineer: Real-Time Model Routing Remote

Kimchi; Netherlands; Remote; EUR 140,000 - 210,000; Full time

Benefits offered by this job

Equity options

Learning budget for development

Hackathons & team events

Equipment budget

Job summary

Kimchi is seeking an experienced platform ML engineer to own the routing model end-to-end, including features, training targets, calibration, and cadence. You will design production-parity evaluations, compare router-vs-router performance, and convert fleet outcomes into training signals with clear failure categorization.

The role emphasizes calibrated, speed-aware decisions. You will lead online experimentation, shadow and full rollout programs, and monitor drift to retrain when needed.

Qualifications

• 5+ years in applied ML or ML research with real-time decision models (ranking, routing, or pricing).
• Strong Python and SQL at scale; ability to interrogate millions of request-level records.
• Experience with LL routing concepts and benchmarks.

Responsibilities

• Own the routing model end to end: features, objective, training, calibration, release cadence.
• Build evaluation gates for production parity and performance comparisons.
• Turn fleet data into training signals with actionable failure categorization.
• Model user-facing objectives: task success × token cost × wall-clock speed under concurrency.
• Run online loops: shadow, A/B, full rollout with drift monitoring and retraining.
• Push frontier with calibrated decisions and abstention under uncertainty.
• Partner with platform engineers on instrumentation and data-path support.

Skills

Applied ML research

Python

SQL at scale

LLM routing

Experimentation & A/B testing

Production ML systems

Calibration methods

Tools

PyTorch

ClickHouse

OpenAI/Anthropic APIs

GitLab CI

Job description

Kimchi is seeking an experienced platform ML engineer to own the routing model end-to-end, including features, training targets, calibration, and cadence. You will design production-parity evaluations, compare router-vs-router performance, and convert fleet outcomes into training signals with clear failure categorization.

The role emphasizes calibrated, speed-aware decisions. You will lead online experimentation, shadow and full rollout programs, and monitor drift to retrain when needed.