AUMOVIO
Multi-agent defect detection for automotive software
Agentic code scanning system that found 30+ high-priority defects in a production automotive codebase.
Nikita Kozodoi
Senior Scientist on the AWS forward-deployed engineering team. Over the past eight years I've finished a PhD in ML and shipped AI models into production across multiple industries. I've found that the most interesting problems require a mix of applied science, engineering and business understanding.
At AWS, I embed with customer engineering teams and tech-lead projects end to end: framing the problem, running experiments, fine-tuning and evaluating models, deploying the result into production. My customers include Nasdaq, Ryanair, adidas and Zalando. My current work focuses on LLM-based and agentic systems.
Alongside the customer projects, I publish applied research papers and blog posts, and speak at conferences and meetups. I'm interested in LLM fine-tuning, inference-time optimization and foundation models for tabular and time series data. I've also spent several years on Kaggle, reaching Competitions Master and building ML solutions across different modalities.
Experience
2022 — present
Senior Applied Scientist
Amazon Web Services · Berlin
- ·Designing, building and deploying custom AI and agentic solutions to solve diverse business problems.
- ·Tech-leading projects for large-scale customers such as Nasdaq, Ryanair and adidas.
- ·Published 6 papers, including NeurIPS and ICML workshops, and open-sourced 4 reusable assets.
- ·Led a squad of 6 scientists and conducted 80+ interviews; promoted from Applied Scientist in 2024.
2022
Project-Based Data Scientist
Simply Rational · Berlin, freelance
- ·Prototyped customer-segment prediction models from incomplete data for 4 finance companies.
- ·Analyzed A/B test results to identify treatments improving customer repayment by up to 6%.
2018 — 2022
Data Science Research Associate
Monedo · Berlin
- ·Developed sample bias mitigation algorithms that improved model profitability by up to 3%.
- ·Reduced data acquisition costs by up to 25% using multi-objective feature selection methods.
Education
2018 — 2022
PhD in Machine Learning
Humboldt University · Berlin
Published 5 papers on ML for credit risk, earning 500+ citations and the EURO Best Paper Award. Degree with Honors.
2015 — 2017
MSc in Economics, Quantitative Methods
Humboldt University · Berlin
Focused on predictive analytics and machine learning. Won awards in 2 data science competitions.
2010 — 2014
BSc in Economics
Higher School of Economics · Moscow
Focused on statistics and econometrics. Ranked top-1 across the 4-year student cohort; degree with Honors.
Selected work
Toyota Motor Europe
Recovering business logic from a mainframe
Agentic pipelines that recovered the business logic buried in 1.3 million lines of mainframe code, now moving to production rollout.
Lounge by Zalando
Localizing marketing content across 17 European markets
Test-time self-reflection for content localization across 17 European markets, tuned for quality, cost and latency.
Writing & Speaking
Open Source
Deep Research with Bedrock AgentCore
Agentic application on Amazon Bedrock AgentCore that automates multi-source deep-research workflows.
Lead developer and owner
LLM Self-Reflection
Python library implementing inference-time self-reflection, voting, and augmentation for Amazon Bedrock models.
Core developer and maintainer
Credentials
Kaggle Competitions Master
18 competition medals and a top-1% ranking across Competitions, Notebooks and Datasets.
EURO Award for Best Paper 2025
Awarded by the Association of European Operational Research Societies for the paper on fair ML in credit scoring.
12 professional certifications
7 AWS certifications including Generative AI Developer and ML Engineer, 3 Anthropic Claude certifications, 2 Udacity nanodegrees.