Cinder Machine Learning Engineer at Cinder building ML systems to improve classification pipelines and detection strategies for harmful content. Responsible for end-to-end model development from data processing to deployment.
Responsibilities
Turn real-world customer data into something a model can actually learn from, then decide what model approach fits: a classical classifier when it wins on cost and latency, a fine-tuned LLM when the tradeoff is worth it, a third-party API as a bootstrap. You own the full path from data to decision, not just the model.
Improve our classification pipeline, confidence cascading, and detection strategies so we catch harmful content efficiently — balancing cost, latency, and accuracy deliberately.
Develop intelligent features that help moderators make decisions, organize platform content, and reveal patterns across our data.
Partner with Engineering to build out Cinder's in-house model training, hosting, and inference platform.
Design and build the evaluation and metrics infrastructure customers rely on, including how classifier scores and model outputs are calculated, stored, surfaced, and iterated on.
Partner with our Founding Data Scientist and AI Engineers to shape the agent evaluation architecture — measuring whether our agent fleet is making the right decisions with the right tools at the right cost.
Partner with our Data Engineer to shape the data infrastructure powering our ML systems, ensuring model training, feature pipelines, and production inference have the right data flowing at the right latency and scale.
Mentor teammates and raise the ML bar across the company as Cinder's ML capability matures.
Qualification
RealDeep fluency with the fundamentalsSolid MLOps foundationLocation & Benefits
Required
5–8+ years of machine learning engineering experience on a small team, with a strong track record of shipping ML systems (gradient boosting, tree-based models, classifiers, embedding-based methods) to production.
You've taken a classification problem from messy, unlabeled, real-world data all the way to a model that shipped and stayed load-bearing in production, not handed off to someone else to deploy, and not just scored against a clean, idealized benchmark dataset
You understand LLMs well enough to make an informed, defensible call about when an LLM is worth its cost and latency versus a classic model.
Real, hands-on experience building classifiers under severe class imbalance, where the signal you care about is a small minority of the data
Thrived in environments where the ML infrastructure wasn't already built for you: you've stood up training pipelines, serving infrastructure, evaluation harnesses, and monitoring from scratch rather than inheriting a mature platform.
Startup or small/mid-size company experience where you owned meaningful scope and had to make pragmatic tradeoffs about what to build, what to buy, and what to defer.
Deep fluency with the fundamentals: thoughtful feature engineering, leak-aware train/test splits, metric selection on imbalanced data (precision/recall/F1/AUC over accuracy), cross-validation, and principled hyperparameter tuning.
Strong Python skills and hands-on experience with AI & ML frame works: (PyTorch, scikit-learn, langchain, XGBoost etc)
Solid MLOps foundation: CI/CD for ML, model versioning, experiment tracking, drift detection, and production monitoring. Bonus if you have experience with training, evaluating and serving models via Databricks
Experience designing inference systems with explicit latency and throughput targets, and independently making informed tradeoffs between model complexity, cost, and performance.
Experience with AWS and infrastructure-as-code (Terraform) is a plus.
