Jinwoo Je | Portfolio

I built an interim UPF forecast system for RSR, the rural last mile network. After organizational merger earlier this year, achieving parity became high priority. The system leverages the same metric flow while utilizing available data sources without additional data engineering. I added customization for RSR- specific use cases.
I focused on building Zipscape, an auxiliary product designed to work with UPF. While UPF handles forecasting and plan artifact generation, detailed data remains behind the scenes in the SageMaker layer, not easily accessible without setting up compute instances. With multiple stakeholder requests for easier, more intuitive zip-level volume visibility, I created Zipscape using a database-less solution: API Gateway, partitioned Parquets in S3, Lambda with DuckDB (backend), React, Midway auth, and CDN (frontend).
I worked deeply with KIRO, an Amazon-developed spec-driven AI development tool. Like others, it's a VS Code fork, but highly opinionated. KIRO follows production-level software development workflow: Requirements (BRD/PRD) → Spec (Technical Document) → Build (backlog, sprint, ticket). This is the first AI development tool I've seen where I can learn best practices. Using KIRO felt like working with capable product managers, tech leads, and coders. The PM and tech lead aspects are especially unique—requirements and technical spec drafting were exceptional. The structured approach ensures requirements, specs, and builds are connected, linked, and traceable. It has a steeper learning curve and requires more steering than other AI coding tools, but I'll keep it in my arsenal, especially for 0-to-1 development.