Skip to content
Almostly Almostly

About

Denis Burakov

I’m Denis Burakov — a Lead Data Scientist & ML Engineer based in Berlin, with 10+ years building high-impact ML in fintech and banking. The tools on Almostly grew out of that work.

Reach out about

Section titled “Reach out about”
  • Fine-tuning LLMs — adapting open models to real tasks: data, training, evaluation, and getting them into production.
  • Machine learning for financial services — credit risk, fraud, and decisioning systems, end to end.

Experience

Section titled “Experience”
  • Renmoney — Data Science Team Lead (2025–present). Leading a team on real-time credit scoring, collections and marketing models, and vision-RAG anti-fraud.
  • Amazon — Senior Risk Manager, Data Science (2024–2025). Built customer-abuse fraud-prevention models (+$20M entitlement) and a GPU FAISS + graph-clustering pipeline grouping millions of multilingual product embeddings into interpretable families; ran ML workloads on AWS (Glue).
  • N26 — Credit Risk Modeling Manager (2021–2024). Led four data scientists; shipped an automated credit-scoring engine that cut net credit losses by 50% (+$12M annualized, +30% RoE); automated IFRS9 reporting on Redshift + dbt.
  • Earlier: KPMG (Risk & Treasury data science) and Sberbank (Head of Data & Analytics, International Business).

Publications & writing

Section titled “Publications & writing”
On Credit — the book My book on credit risk modeling: the theory and craft behind scorecards and profitable lending decisions.
An Information-Theoretic Framework for Credit Risk Modeling arXiv:2509.09855 — unifies Weight of Evidence, Information Value, and PSI as classical information divergences, with standard errors and a principled performance–fairness frontier for interpretable scorecards.
Beyond credit scoring: a framework for profitable lending Taktile — an NPV framework that moves lenders past the credit score to optimize profitability across pricing, cost of risk, and acquisition cost.

On Credit is also available on Amazon.

Profiles & affiliations

Section titled “Profiles & affiliations”
Taktile — author profile Risk modeling and data science expert; contributing author on lending and financial decision-making.
Credit Research Centre, University of Edinburgh External Affiliate of the Credit Research Centre at Edinburgh Business School.

Open source

Section titled “Open source”
xBooster Explainable boosted scorecards from XGBoost, LightGBM, and CatBoost.
fastwoe Fast Weight-of-Evidence (WoE) encoding and inference.

Plus wrencode — a minimal agentic coding assistant.

Credentials

Section titled “Credentials”
AWS Certified Generative AI Developer – ProfessionalAWS Certified Data Engineer – AssociateAWS Certified Machine Learning Engineer – AssociateAWS Certified AI Practitioner

AWS Certified — Generative AI Developer (Professional), Data Engineer (Associate), Machine Learning Engineer (Associate), and AI Practitioner. All badges on Credly →

Fellow of the Royal Statistical Society · M.A., Social Science / Statistics, University of Denver.

Get in touch

Section titled “Get in touch”