Nicholas Stamatakis

Software Development Engineer Intern

Serverless data pipeline expanding coverage 27× with 95% ML accuracy.

• Designed a serverless web-scraping pipeline with Scrapy on AWS Lambda, expanding coverage by ~27×.
• Integrated Claude 3.7 Sonnet via AWS Bedrock for NLP, yielding 95.35% accuracy on backtested datasets.
• Automated infrastructure deployment with AWS CDK, managing 100K+ JSON files in S3.
• Orchestrated a Step Function state machine with integration tests to validate the end-to-end CI/CD pipeline.
• Instrumented 10+ CloudWatch dashboards monitoring program health and error metrics.

Software Engineer Intern

AI exam prep platform with 75,000+ users.

• Launched whiz.study, a full-stack preparation platform serving 75,000+ users using a TurboRepo monorepo with Next.js, Expo (React Native), and Express across 6 apps with a 93-model PostgreSQL schema via Prisma ORM.
• Built an agentic AI tutor using Vercel AI SDK, OpenRouter, and a Bun WebSocket server.
• Engineered an adaptive study plan engine ("Road to 1600") that builds personalized lesson sequences using topological sorting, all as pure functions with 1,000+ automated tests.

Software Engineer Intern

• Engineered sentiment analysis system using Large Language Models, processing 200+ monthly feedback entries.
• Implemented semantic search over 8K+ JIRA tickets using Retrieval-Augmented Generation (RAG).
• Optimized a Qdrant Vector DB for efficient storage of System Problem Reports; reduced data retrieval time by 70% and enhanced report generation for 20,000 performance entries.

Adversarial ML Researcher

Published research: auditing 10,000 ballots with 99.5% accuracy.

• Designed a support vector machine to audit 10,000 ballots for the state of Connecticut, achieving 99.5% validation accuracy.
• Implemented adversarial machine learning techniques to perform attacks on the model, reaching a detection rate of 98% for adversarial examples.
• Constructed advanced image registration software using Python and OpenCV; aligned 10,000 printed and scanned ballots, improving accuracy and reducing manual verification time by 85%.

Data Structures & Algorithms Teaching Assistant

• Created 50+ practice problems, led exam reviews, and proctored for 200+ students. Led weekly 1-hour recitations.