Data Scientist & Software Engineer | ETH Zürich MSc | Building Production ML/Software Systems at Scale
Explore My WorkData Scientist & Software Engineer with expertise in building production-grade ML systems and scalable distributed software. Currently pursuing MSc in Data Science at ETH Zürich while conducting research on diffusion language models at IBM Research. Strong foundation in systems engineering, quantitative methods, and applied machine learning from experience at leading tech and finance firms.
Specialized in high-performance computing, low-latency systems, and end-to-end ML pipelines spanning LLMs, agentic workflows, and quantitative analysis. Winner of 200+ participant Optiver trading challenge and contributor to Fortune 500 partnerships. Passionate about building robust, scalable systems that bridge theoretical research with production deployment.
2+ Years in AI and Software Engineering
Semantic Segmentation and Depth Estimation project for autonomous vehicles. Implemented DeepLabV3+ with multi-task learning for dense prediction tasks including semantic segmentation and depth estimation.
Built a 2-stage object detector for autonomous vehicles using LiDAR point clouds. Implemented Region Proposal Network and Refinement Network for coarse and refined detections, with multi-modal data visualization including 3D bounding boxes and semantic segmentation.
Production-scale computer vision pipeline for Fortune 500 aviation client (Lufthansa). Built end-to-end anomaly detection system processing 10k+ aircraft images, achieving 95%+ accuracy. Deployed scalable MLOps pipeline with automated retraining and monitoring.
Advanced research on Large Language Model security, implementing watermarking schemes, attack methods, and content filtering using CLIP embeddings. Published-quality research addressing critical AI safety challenges in production LLM systems.
Comprehensive study on Large Language Model calibration using Microsoft Phi-2 and GSM8K dataset. Implemented various prompting strategies including Chain-of-Thought, subquestion decomposition, and self-consistency approaches with temperature sampling.
Built a question-answering system using RAG with Google's FLAN-T5-small model. Demonstrated significant performance improvements by providing relevant context, creating an adaptable QA system that updates without retraining.
Enterprise-grade RAG pipeline processing 50k+ legal documents for professional legal research. Built hybrid retrieval system with ColBERT embeddings, Elasticsearch, and advanced query routing. Implemented novel LLM-as-judge evaluation framework achieving 90%+ accuracy on domain-specific queries.
Machine learning system to predict student dropout risk using comprehensive data science pipeline with feature engineering, model selection, and deployment using various ML frameworks.
Built predictive ML models using Gaussian Process regression on Europe-wide dataset for air pollution forecasting (PM2.5), achieving superior accuracy for environmental health applications.
Comprehensive implementation of big data technologies including S3, DynamoDB, HDFS, and Spark for scalable data processing and analytics on large-scale distributed systems.
Developed intelligent chatbot enhanced with university website data using OpenAI, LangChain, and Pinecone for question-answering and information retrieval with web scraping capabilities.
Developed a fully automated data pipeline for collection of tokenomics and whitepapers for up to 10,000 cryptocurrencies, utilizing web scraping and multiple API integrations. Built scalable infrastructure for real-time crypto data processing and analysis.
Replaced obsolete keyword-based whitepaper analysis with an advanced pipeline using Large Language Models for in-depth automated whitepaper analysis. Implemented sophisticated NLP techniques for extracting insights from cryptocurrency project documentation at scale.
Won 200-participant hackathon by building linear regression model with advanced feature engineering for option pricing. Demonstrates quantitative finance expertise and competitive programming skills.