Senior Data Scientist

About Akerio. Akerio is a UK-based applied AI company. We build platforms that turn messy enterprise data into scored, auditable recommendations: the next diesel price for one of Europe's largest independent fuel distributors, a risk flag on a tenant across 41 shopping centres, a credit score a bank's committee can trace back to the source record. Our models are built by a PhD team from Stanford and the University of Chicago, every output shows what drove it and how confident it is, and the platform runs on-premise or in the cloud. We are live in production across energy, commercial real estate and financial services, with 18 AI modules running in one deployment alone. We are a small team shipping to demanding enterprise clients, so the work you do reaches production and gets used. • Job: Full time • Job type: Remote The role Our clients do not buy a model. They buy a better decision: which lease to renegotiate this quarter, which tenant is about to pay late, how much footfall a mall will see next month. On our retail property platform, revenue, footfall, tenant risk and mall performance models already run every month through Airflow across 41 shopping centres. The next phase adds collections and cash projections, campaign ROI, and models trained per client. You own that model portfolio: accuracy, explainability, retraining, and the step from "runs on one client's data" to "runs for every client". You work with our Chief AI Engineer and directly with the client's asset, leasing and finance teams. What you will work on • Forecasting: tenant turnover, hourly footfall per location over a 12-month horizon, energy use, collections and cash flow. • Risk scoring: tenant payment-delay risk, brand risk scores, early warnings on underperforming units. • Lease analytics: estimated rental value (ERV) and which contracts sit below market. • Explainability: every output ships with drivers (SHAP) and a confidence range a leasing manager can understand. • MLOps: moving from scheduled scripts to proper experiment tracking, a model registry and per-client training, with clear accuracy reporting back to the product. • LLM quality: the AI assistant answers questions over these numbers with 53 tools. You help design its evaluation: test sets, terminology rules, and the Langfuse feedback loop. Beyond real estate, the same skills apply to our energy pricing and banking risk work, so you will see more than one industry. Our stack Area What we use Modelling Python, pandas, scikit-learn, XGBoost, Prophet and other time-series methods Explainability SHAP Orchestration and serving Apache Airflow for monthly training and scoring runs, FastAPI compute service, MLflow being introduced Data PostgreSQL and PostGIS, Excel and ERP exports from client systems LLM Anthropic Claude via API, RAG over client data, tool calling and MCP, Langfuse for conversation and quality monitoring Infra Azure, Docker, Jenkins, Bitbucket How we build Jira, spec-first work with AI coding tools (Claude Code), code review on every PR Your first six months • First month: you know one live model inside out (data, features, metrics, failure modes) and ship one measurable improvement to it. • Month three: you own a model family (for example tenant risk) including retraining, monitoring and the accuracy numbers clients see. • Month six: you lead the modelling side of a new client rollout or new module, from use-case scoping to an advisory-mode pilot, and mentor a junior. What you bring • 5+ years building ML models that ran in production and changed real decisions, not only research or dashboards. • An MSc or PhD in a quantitative field (statistics, computer science, economics, operations research), or equivalent depth. • Strong time-series forecasting, plus at least one of: risk or default scoring, demand forecasting, pricing. • Sound statistics: you know when a backtest lies, how to handle sparse or messy client data, and how to explain uncertainty to a CFO. • Production Python: clean, tested, reviewable code an engineer can deploy without rewriting it. • Experience putting models on a schedule (Airflow or similar) and tracking them (MLflow or similar). • Fluent written and spoken English. Nice to have • Retail, shopping centre, real estate or energy data. • LLM evaluation in production: test sets, LLM-as-judge, failure analysis from real user sessions. • Geospatial data or footfall and sensor data. • Multi-tenant model training: one pipeline, many clients, different data quality. • Turkish, helpful for working with current client teams. How we work Every model starts in advisory mode next to the client's current process and earns trust by beating it on agreed metrics. You present results to client teams yourself, so you hear quickly whether a model helps. Hiring process 10. 30-minute intro call. 11. 30-minute session: present your case to the CPO and an engineer. 12. 15-minute call with a founder. What we offer • Fully remote, with overlap on UK and Türkiye working hours. • Models in production with real outcome data coming back every month. • A PhD-heavy AI team to think with.