About the Role
InPost Pay is building its own fintech platform from the ground up. It's a large-scale, international initiative where engineers will have the opportunity to shape the architecture, build core fintech components, and create solutions designed to operate at InPost's scale.
Responsibilities
•Own agentic model iteration (experiment, evaluate, retrain, promote).
•Own agentic model monitoring and observability (drift, stability, performance), built on the feature store.
•Design the agent loop and graph and keep it reliable and cost-controlled.
•Own agentic validation and backtesting. Build agent loops that replicate and stress-test models and produce evidence packs for independent review.
•Own agentic model monitoring and observability for validation.
•Build and ship credit-scoring and merchant-underwriting models end-to-end: feature engineering, training, validation, deployment and post-production monitoring.
•Engineer point-in-time-correct features against the shared feature store for BNPL and MCA decisioning.
•Partner with Data Engineering on source contracts and freshness SLAs.
•Design and operate the agent loop and graph (context, tools, evals, guardrails) so agentic delivery is reliable, observable and cost-controlled.
•Deploy and operate models with MLOps tooling: versioning, CI/CD for models, and champion/challenger or shadow deployment.
•Monitor production models for drift, stability and performance degradation.
•Close the loop with retraining, recalibration or rollback when thresholds are breached.
•Design models to be explainable and regulatory-defensible by construction.
•Document features, decisions and validation evidence to support CCD2 creditworthiness assessment and audit review.
•Model Validation track: build and run the validation and backtesting harness inside the MLOps pipeline. Produce audit-ready evidence packs for independent review.
•Model Lifecycle track: build the experiment, evaluate, retrain and promote loop on the model registry and serving layer.
Requirements
•Someone who loves building ML systems that hold up in production and under audit.
•Hands-on first. You design and you build, with AI as a daily tool.
•Experienced in applied ML at scale, covering the end-to-end lifecycle: feature engineering, training, validation, deployment and monitoring.
•Skilled in model validation methodology: backtesting, population stability (PSI), discrimination (Gini/KS), calibration.
•Experienced in MLOps and production monitoring: model versioning, CI/CD, champion/challenger and shadow deployment, drift detection and automated alerting.
•Skilled in point-in-time-correct feature engineering and feature-store consumption patterns.
•Familiar with explainability techniques (SHAP/LIME) and documentation for regulatory defensibility.
•Fluent in Python and SQL, with Spark/Databricks for large-scale model training and serving.
•Experienced with advanced agentic tooling and loop and graph engineering: context and tool design, guardrails, eval design with auditable outputs.
•Solid working knowledge of LLMs.
•A collaborator who follows shared patterns and raises the bar through influence.
•Fluent in English (B2+).
Nice to Have
•Hands-on model validation or independent-review experience.
•CCD2 / consumer-credit regulatory requirements for creditworthiness assessment.
•Merchant cash advance (MCA) underwriting and merchant-risk data.
•Fraud-detection modelling (e.g. bust-out, first-party fraud).
•IFRS 9 provisioning / expected-credit-loss modelling.
•Real-time, low-latency model serving.
Conditions
•Greenfield, high stakes.
•Strategic impact on how AI-native ML is done across InPost.
•A modern stack and a lean, AI-native team.
•Space to experiment with new technologies, including AI tooling, and bring them into production.
•A collaborative, fast-moving environment where autonomy and accountability go hand in hand.
•Hybrid, ideally one day a week in the Warsaw or Kraków office.
•B2B Contract.
