Senior Machine Learning Engineer (EU), Scoring & Audience Models

We're hiring a Senior Machine Learning Engineer to build the models that decide who our customers should target — account scoring and company lookalikes over messy, real-world B2B data. The product is live and used by thousands of customers; your job is to make targeting accurate enough that they trust it by default. This is a high-ownership modeling role, and an early one. Our product runs on Node.js, React and Elasticsearch — the modeling layer is yours to define. You'll own the problem end to end: the labels, the features, the model, how scores get served, and what happens to all of it in production six months later. What You'll Do Own account and lead scoring end to end — target definition, label design, feature engineering, model selection, and the calibration that makes a score mean something Build lookalike and similarity systems that take a customer's best accounts and find the next 500, and prove they beat a well-tuned baseline before they ship Work Elasticsearch hard as both feature source and serving layer — aggregations over company and engagement documents, vector search for candidate retrieval, and model scores indexed back as ranking signals Design evaluation that holds up — validation free of temporal and account leakage, metrics matched to the decision (lift and precision at k, calibration), and online experiments with real power Fix the data underneath the model: entity resolution, deduplication, missingness, label noise, sampling bias, and the feedback loops that appear once a score starts changing user behavior Stand up the training and inference path — pipelines, feature storage, batch and near-real-time scoring — and integrate it cleanly with our Node.js services Monitor production models for drift and degradation, and retrain on a defensible cadence Work with product and engineering to surface scores in the application, and explain to customers what a score does and doesn't mean What We're Looking For Must have: 4+ years in machine learning or data science roles, including 2+ years shipping supervised models that drove real decisions in production Formal grounding in machine learning, deep learning, and advanced statistics — a quantitative degree or equivalent rigorous training. You're comfortable with the mathematics, not just the APIs Deep expertise in applied statistics: estimation and inference, sampling and selection bias, hypothesis testing and power, regularization and bias/variance, probability calibration, causal versus correlational claims Strong command of the modeling stack — Python, scikit-learn, XGBoost/LightGBM, PyTorch or TensorFlow, pandas/Polars, NumPy, SciPy Fluency extracting and aggregating features from large document stores at scale. We run Elasticsearch; equivalent experience with another NoSQL or search-based store transfers, as does strong SQL, but you should be ready to work in ES query DSL and aggregations directly Experience with ranking, propensity, similarity, recommendation, or churn/conversion modeling, where the output is a score that changes what someone does A model you can defend in detail: why that target, where the labels came from, what your validation missed the first time, what broke after launch Comfort building the ML foundations rather than inheriting them — you've been early somewhere and know which infrastructure is worth building in month one versus month twelve Enough engineering discipline to own your pipelines end to end — version control, testing, CI, containers — and to hand a scoring service to a Node.js team without drama Nice to have: Elasticsearch depth — dense_vector and kNN search, significant_terms and other statistical aggregations, rank_feature scoring, index design and query performance tuning at scale B2B data, GTM, martech, adtech, or CRM enrichment background, and familiarity with firmographic data providers Approximate nearest neighbour search beyond ES (pgvector, Qdrant, FAISS) with a real grasp of the recall/latency trade-offs Learning-to-rank in a search context Graph or network modeling over company and contact relationships MLOps tooling — Airflow or Dagster, MLflow or Weights & Biases, dbt, Feast Publications, competition track record, or open-source contributions in applied ML P.S: this is not an LLM application role. RAG, prompt engineering, and agent orchestration are adjacent skills we're glad to see, but they don't substitute for the modeling foundation above. If your ML experience is primarily building on top of model APIs, this particular role isn't the right fit — we'd rather say so now than at the technical screen.