Career

Experience

From economics and consulting to production ML and agentic AI at Novartis.

Download CV

Work experience

Senior AI Expert

Current

Novartis · leading the agentic platform since Apr 2026

Oct 2026 – PresentBarcelona, Spain

Build the backbone architecture and shared platform for agentic applications: design patterns for agent-to-agent communication, reusable Python packages, integrations with internal systems and CI/CD templates that let teams build and ship new agents consistently and faster.
Design and develop multi-agent systems in LangGraph on AWS for complex finance workflows, built on orchestrator–subagent patterns with semantic model definitions over SQL data, reusable specialized subagents (iterative knowledge retrieval, SQL-based research, dynamic visualization) and human-in-the-loop controls.
Took a complex finance agent to production over 100M+ data points across 100+ tables, holding the hallucination rate under 1% across live user interactions and tester sessions.
Develop the agentic evaluation harness used across every agentic application we build, combining LLM-as-a-judge and deterministic metrics with fixture-based, ground-truth and LLM-generated synthetic test sets, so every change to architecture, models or context is validated before release and ships only when it delivers a measurable improvement.
Ship full-stack agentic applications with React and TypeScript frontends and FastAPI services on AWS (Lambda, ECR, S3, DynamoDB), using hallucination guardrails, tracing, staged fallbacks and continuous feedback loops to keep agents reliable and self-improving.
Onboard and train engineers and analysts on agentic AI patterns, the shared platform and its architecture standards.

PythonTypeScriptReactAgentic AIMulti-Agent OrchestrationLangGraphLangChainOpenAI Agents SDKAnthropic Claude SDKFunction Calling / Tool Use

Expert Data Scientist

Novartis

Jun 2025 – Oct 2026Barcelona, Spain

Replaced a quarterly, Excel-based finance forecast with daily automated ML forecasts covering €20B+ in yearly costs across 30+ countries — now the baseline for analysts and senior finance managers.
Cut per-country quarterly MAPE to under 1% — 50%+ better than analyst forecasts, with significantly larger gains at finer granularity — saving 150+ analyst hours a month.
Product owner of the gross-to-net and tech-ops forecasting products: took them from POC to production and ran the roadmap, stakeholder alignment and quarterly business reviews.
Shipped agentic products into finance workflows: multi-agent systems on AWS Bedrock and OpenAI Agents SDK, function-calling agents over Snowflake/Databricks, RAG over policy and contract corpora, MCP servers linking LLMs to forecasting, expense analysis and reporting, and human-in-the-loop confirmation for sensitive actions.
Owned LLM evaluation and rollout: model selection across Bedrock, OpenAI, Anthropic and local Ollama deployments, Langfuse tracing and compliance review with leadership before production cutover.
Took 4+ core ML and AI products to production, moving forecasting to millions of daily data points with ensemble models, rolling cross-validation, backtesting and monitoring.

PythonSQLPandasPolarsNumPySciPyPySparkScikit-learnCatBoostXGBoost

Data Scientist

Novartis

Sep 2023 – Jun 2025Barcelona, Spain

Built the MLOps foundations for Dockerized data-science projects: Jenkins CI/CD on AWS, a scalable model registry for reproducible deployment and automated unit, integration and end-to-end testing of ML workflows.
Improved ML models for time-series forecasting, A/B testing and cost and sales prediction, and developed the forecasting methodology: outlier detection, model selection, backtesting and performance evaluation.
Built automated, interactive dashboards in Power BI, Streamlit and QlikView for real-time insights.

PythonSQLPandasNumPyScikit-learnCatBoostXGBoostLightGBMTime Series ForecastingAnomaly Detection

Data Scientist

Citibeats

Mar 2023 – Jul 2023Barcelona, Spain

Developed NLP pipelines for large-scale text classification and sentiment analysis, processing millions of social media posts to identify public opinion trends for institutional clients (WHO, UNDP).
Built and fine-tuned Hugging Face transformer models (BERT, Llama) for multilingual text classification and topic modeling across 10+ languages.
Implemented unsupervised ML techniques (clustering, topic extraction) on unstructured text data to detect emerging social narratives and community needs.
Worked with Twitter API for real-time data ingestion and built ETL pipelines for large-scale text preprocessing and feature extraction.
Contributed to the team's 30+ fairness and robustness test suite ensuring NLP model quality and bias mitigation.

PythonNLPTransformersHugging FacePyTorchScikit-learnPandasNumPyFeature EngineeringTwitter API

Teaching Assistant

National Research University Higher School of Economics

2019 – 2022Saint Petersburg, Russia

Led weekly seminars in Mathematical Statistics and Microeconomics for groups of 30+ undergraduate students across 3 academic years.
Designed problem sets, graded examinations, and provided individual tutoring for students struggling with quantitative methods.
Facilitated Q&A sessions bridging theoretical concepts (game theory, probability, optimization) with real-world economic applications.
Received consistently high student evaluation scores; helped improve average exam pass rates in assigned sections.

MicroeconomicsMacroeconomicsStatisticsEconometricsGame TheoryBayesian StatisticsPythonRStataLaTeX

Consultant

PricewaterhouseCoopers (PwC)

Jan 2021 – Jun 2021Saint Petersburg, Russia

Performed transfer pricing analysis for multinational clients: benchmarking financial performance, identifying comparable transactions, and evaluating intercompany pricing policies.
Analyzed global corporate structures and financial statements across 10+ jurisdictions to assess compliance with OECD transfer pricing guidelines.
Prepared transfer pricing documentation and economic analyses supporting clients' tax positions for regulatory review.
Collaborated with cross-functional teams (legal, tax, audit) on complex restructuring and M&A due diligence engagements.

Financial ModelingMicroeconomicsMacroeconomicsStatisticsEconometricsExcelPower BI (DAX, Power Query)UI/UXLaTeXSQL

Education

Master's Degree in Data Science

Barcelona School of Economics

Thesis: Early Prediction of Preeclampsia from cf-RNA with ML

★ Top 20% of class

2022–2023Barcelona, Spain

Bachelor's Degree in Economics

National Research University Higher School of Economics

Thesis: Statistical analysis of excess mortality factors in Russia during COVID-19

★ Top 10% of class

2018–2022Saint Petersburg, Russia

Erasmus Program in Economics

University College London (UCL)

★ Erasmus+ Scholarship

2020–2021London, UK

International Foundation Program

University of Toronto

★ Award of Excellence

2017–2018Toronto, Canada

Piscine — Intensive C Programming

Codam Coding College

2019Amsterdam, Netherlands