Stuti Nayak

ML Engineer · Data Engineer

Stuti Nayak

Portrait of Stuti Nayak

4+ years building production ML systems, NLP and speech classifiers, and LLM-powered applications on real-world data pipelines and cloud infrastructure.

The Hague, Netherlands Open to roles NL / EU

What I Build

  • Machine Learning Systems NLP and speech classifiers (BERT, RoBERTa, wav2vec 2.0)
  • LLM Applications Retrieval-Augmented Generation systems
  • Data Infrastructure Event and telemetry pipelines, data quality and ML deployment

Selected Systems

Filter by skill

Food QA Retrieval System

View →

Production-style Q&A API using Retrieval-Augmented Generation with document chunking and vector indexing.

Python FastAPI LangChain RAG Transformers

Credit Card Fraud Detection

LightGBM classifier on 284,807 highly imbalanced transactions, with temporal splitting, threshold optimisation and SHAP analysis of model behaviour. ROC-AUC around 0.98.

Python LightGBM SHAP

NLP Classification Pipeline

View →

BERT/BERTopic multi-label classification; RoBERTa sentiment analysis pipeline deployed on AWS.

Python PyTorch Transformers AWS

Streaming Data Pipeline

IoT simulation pipeline using Kafka ingestion, processing layer, and ML analytics on Google Cloud.

Python Kafka MLFlow GCP

Churn Prediction

View →

End-to-end ML project predicting customer churn for a telecom company with EDA and model deployment.

Python Scikit-learn Pandas SQL

Experience

  1. Data Engineer, Aurelius Enterprise

    Amsterdam · – Present

    Production event and telemetry ingestion pipelines that turn noisy upstream data into schema-validated datasets. Data quality rules, dataset contracts, lineage and governance with client and engineering teams. Evaluation criteria for a client-facing AI automation initiative, and contributions to an open-source metadata platform.

  2. Data Scientist, Knownwell

    The Hague · –

    BERT / BERTopic classification and topic-routing model, improving match accuracy from 20% to 30% on real-world text. RoBERTa sentiment classifier validated with held-out evaluation and error analysis.

  3. Data Scientist, Rate.nl

    Rotterdam · –

    wav2vec 2.0 audio classifier on noisy call recordings, improving accuracy from 27% to 55%. Experiments tracked in MLflow, containerised inference with Docker on Google Cloud Run, and a Gradio app for stakeholder review.

  4. Researcher, Clinical Data Science, Maastricht University

    Maastricht · –

    Federated learning infrastructure (Vantage) for privacy-preserving cross-hospital analytics; semantic model to standardise data across hospitals.

  5. Research Assistant, Institute of Data Science, Maastricht University

    Maastricht · –

    Deep learning NLP pipeline over scientific abstracts for biomedical metadata quality; CNN and BiLSTM multi-label classifiers, with the CNN reaching 83% accuracy.

Toolkit

ML / AI
PythonPyTorchTensorFlowHugging Face TransformersBERT / RoBERTawav2vec 2.0scikit-learn
LLM / Retrieval
RAGLangChainFAISSSemantic search
Evaluation
Held-out & cross-validationPrecision / recallImbalanced learningError analysisMLflow
Data
SQLPandasNumPyPySparkKafkaBigQueryPostgreSQL
Production
DockerGitGitHub ActionsCI/CDFastAPIGradio
Cloud
Google Cloud (Cloud Run, BigQuery)AzureAWS

Education

  • M.Sc. Data Science in Decision Making

    Maastricht University, Netherlands

  • B.Tech. Information and Communication Technology

    Dhirubhai Ambani University, India

Certifications

  • Confluent Certified Developer for Apache Kafka

    Confluent

  • Certified Data Management Professional (CDMP), Associate

    DAMA International

  • Databricks Certified Data Engineer Associate

    In progress

Outside Work

  • Anime fan: always on the lookout for the next great series
  • Fitness enthusiast: staying active and pushing limits
  • Coffee geek: passionate about good brews and the ritual behind them
  • Horror movie lover: the scarier, the better

Contact

I'd love to chat. Reach me over email or LinkedIn about a role, a project, or just ML.