Hello, I'm
Data Scientist with 3 years of hands-on experience in modern data architecture, Lakehouse/Databricks, ETL migration, and production-grade deep learning (NLP/CV). From PySpark pipelines to LLM-powered agents — I build systems that deliver measurable impact.
Data Scientist & Engineer based in Potsdam, Germany — building scalable data platforms, production ML systems, and modern web applications.
3 years at Envestnet Inc. and Cognizant — data pipeline modernization, Lakehouse architecture, anomaly detection, and full-stack development.
M.Sc. Data Science — University of Europe for Applied Sciences, Potsdam.
B.Tech EEE — Govt. Engineering College, Barton Hill, India.
PySpark & Delta Lake pipelines, PyTorch/TensorFlow for NLP & CV, RAG systems with LangChain, and React/Next.js frontends.
Replaced TF-IDF baseline with fine-tuned BERT for mixed stack traces and free text, improving F1 from 0.78 to 0.92. Deployed on AWS EKS with p95 latency under 100ms for live Jira ticket processing.
End-to-end streaming pipeline for financial transaction events with Kafka, Structured Streaming, and Delta Lake sink. Exactly-once semantics with under 30s end-to-end latency at 10k+ events/min.
Fine-tuned ResNet50 for surface defect classification (scratches, cracks) with 96.5% validation accuracy. Achieved 30 FPS on CPU-limited edge devices via INT8 quantization and ONNX Runtime.
RAG system for 5,000+ pages of financial documents with hybrid retrieval (vector + BM25), metadata filtering, and conversation memory. Reduced hallucination rate by 30% on compliance queries.
Fine-tuned BERT-Base for financial document classification across 8 categories with Macro-F1 of 0.91. Optimized to 45ms p95 latency via ONNX INT8 quantization for CPU deployment.
Configurable data quality framework with Great Expectations for automated validation across Bronze/Silver/Gold layers. Lineage tracking with Airflow XComs and dbt manifest parsing visualized in Streamlit.
University of Europe for Applied Sciences, Potsdam
Master thesis: Predictive Maintenance on heavy-duty truck sensor data. Built hybrid ensemble (TFT + XGBoost) outperforming RNN baselines by 18% F1 on 1.1M measurements. Derived EUR 450K annual savings potential via asymmetric loss optimization. Used SHAP for interpretable fault driver identification.
Envestnet Inc.
Refactored legacy stored procedures to PySpark jobs (10M+ rows), accelerating runtimes by 60%. Built Delta Lake pipelines (Bronze/Silver/Gold) for 500k+ daily financial records. Reduced pipeline downtime 75% via Airflow orchestration. Developed LSTM autoencoder anomaly detection reducing server downtime 15%. Deployed inference services with Docker/TF Serving at p99 < 50ms.
Cognizant Technology Solutions
Migrated 10-year-old on-prem Hadoop pipelines to cloud Lakehouse architecture (TB-scale), reducing data availability latency by 40%. Cut cloud compute costs 35% for credit risk scoring pipelines via execution plan optimization. Implemented RBAC and column-level PII masking for zero-trust compliance.
Govt. Engineering College, Barton Hill, India
Capstone: Autonomous vehicle with CNNs and sensor fusion — achieved 30 FPS inference on CPU, reduced steering MAE by 15% vs PilotNet, 95% collision avoidance over 200 test runs. CGPA: 8.0/10.
I'm always open to new opportunities — whether it's data engineering, ML projects, agentic AI systems, or creative frontend work. Drop me a message and let's build something great.