Eric Kauffman

When I joined this project, the platform was already calculating scores for millions of assessments, but it required a major overhaul to meet modern scalability, security, and feature demands.

I led the transformation of this high-stakes scoring system—used by programs like GRE, TOEFL, AP, and SAT—enabling it to process 8+ billion scores per year with 100% accuracy. My role encompassed technical leadership and hands-on development across a team of 10 engineers.

As part of this effort, I containerized the application's core components using Docker and Amazon ECS/AWS Fargate, reducing production costs by 45% while improving resilience and scalability. I also implemented CI/CD pipelines with GitLab, streamlining releases and minimizing manual deployment efforts.

Beyond infrastructure improvements, I developed features for new item types and program-specific workflows and introduced real-time score analytics tools to enhance insights and reporting. Additionally, I spearheaded major security hardening initiatives, significantly reducing vulnerabilities across the stack.

I designed and built a proof-of-concept AI chatbot that enabled users to configure complex scoring requirements through natural-language conversations. This transformed a manual, error-prone process into an intuitive and interactive experience.

Powered by GPT-3.5-Turbo and hosted in Azure to meet enterprise data privacy requirements, the chatbot used LangChain to orchestrate prompt handling, structure user inputs into JSON, and dynamically trigger back-end REST APIs to update scoring configurations in real time.

What made this project exciting was its seamless integration of modern LLM capabilities with existing enterprise architecture—not just as a demo, but as a functional prototype ready for production workflows. Developed and delivered in under two months, it successfully demonstrated how conversational AI could reduce onboarding time, minimize miscommunication, and simplify a traditionally technical setup process.

As part of our broader modernization effort, I spearheaded the design and development of a real-time score reporting service, extracting it from a legacy monolith and transforming it into a dedicated, cloud-native microservice hosted on AWS.

The goal was to reduce reporting latency, and the new service achieved a 10x speed improvement. Additionally, I built an ETL pipeline using AWS Glue and Apache Spark to enhance the legacy bulk score reporting process, increasing throughput by 60%.

This project demonstrated how targeted service decomposition and cloud-native design can drive significant performance gains in large-scale data workflows.