Data Science Lead

Hands-on Data Science Lead on a new engagement with a regulated UK & Ireland credit and lending company. The client has consolidated data from multiple business entities into a newly centralized, anonymized data lake and wants to turn it into validated risk analytics — delinquency, probability of default, credit-policy insight — plus an executive-facing natural-language insight layer.

This is a foundational data-science build, not an agentic-AI project. The early work is unglamorous and hands-on: validating data nobody can yet vouch for, then building defensible models on top. You are the senior data scientist the client is missing — you do the work and own the methodology, while leading a small pod and acting as the human-in-the-loop the client explicitly asked for.

Stage: pre-contract / scoping (Phase 1 = current-state assessment + data validation). Duration: multi-phase, multi-quarter ambition with strong extension probability.

Reporting: Engagement lead / CTO (@Alex Honchar); leads the pod's Data Engineer(s) and the client's offshore data team.

Full-time engagement is preferable.

• Profile the anonymized lake hands-on — interrogate tens-of-millions-of-row tables and reproduce and validate the team's existing descriptive statistics, so every number is traceable to source (the client cannot currently answer “how do you know that's correct?”).

• Build and validate the core risk models yourself: PD, delinquency / roll-rate, early-warning, segmentation and scorecards (WOE / IV, logistic regression, gradient boosting).

• Stand up the model-validation discipline that makes outputs audit-defensible: train / test / out-of-time splits, Gini / AUC / KS, calibration, stability (PSI), backtesting and full model documentation.

• Define feature logic with the Data Engineer and write it yourself in SQL / dbt / Python; specify the harmonized definitions the semantic layer must serve.

• Prototype and validate the natural-language insight layer (text-to-SQL / RAG over the semantic layer); check answer correctness and add guardrails.

• Run a credit-policy / cut-off analysis showing where the client could tighten policy or reduce delinquency — the concrete insight their own clients keep asking for.

• Lead a small pod (Data Engineer, client's junior offshore data people): set tasks, review work, be the quality bar and the human-in-the-loop.

• Front the client's data leadership: present findings, explain methodology to non-technical executives, and shape the phased roadmap / SoW.

• Expert Python for data science (pandas / Polars, scikit-learn, statsmodels) and strong SQL over large tables

• Credit-risk / financial modeling: scorecards, PD, delinquency, segmentation, model validation and governance

• Data validation, profiling and feature engineering on messy enterprise data

• dbt / semantic modeling; partnering with data engineering on the harmonization layer

• GenAI insight layer: text-to-SQL, RAG over structured data, evaluation and guardrails

• Methodology, lineage and documentation that survives audit; able to explain it to executives

• Leadership of small delivery pods and distributed / offshore teams

• GDPR fundamentals (anonymization vs pseudonymization, UK / EU data residency)

• AWS analytics stack and Well-Architected (Analytics, Security) for BFSI

• UK / EU credit & lending regulatory context (FCA, model governance, fair-lending / explainability) — strong plus

• Familiarity with credit-bureau / scoring data products — strong plus

Key characteristics (ideally 4/4):

• Hands-on data science at enterprise scale

• Worked with financial-services / credit clients or in-house at a credit / lending company

• Cloud hyperscaler experience (AWS preferred)

• Technology consulting / client-facing delivery background

Role-specific characteristics:

• 7+ years hands-on data science, with real credit-risk / financial modeling

• Experience building and validating models in a regulated, audited context

• Led small data-science teams while still coding personally

• Demonstrably comfortable doing the data-cleaning grunt work themselves, not just directing it